The Most Cited Science Papers of All Time, Explained

Most people associate scientific fame with dramatic discoveries — the double helix, the theory of relativity, the first image of a black hole. But the papers that scientists themselves cite most often tell a very different story. They are not always the splashiest breakthroughs. They are, more often than not, the tools. The methods. The quiet infrastructure that makes everything else possible.

In April 2025, Nature published a landmark ranking of academic publications by total citation count, drawing on decades of database records. The results are surprising, instructive, and worth understanding — because the most cited science papers of all time reveal something fundamental about how knowledge actually builds.

Here, we dig into the top five most cited papers across five major scientific disciplines, exploring not just what they say, but why they matter and what their enduring relevance tells us about the nature of scientific progress.

What Citation Counts Actually Tell Us

Before diving into the papers themselves, it is worth pausing on what a citation actually means. When a researcher publishes a new study, they list every prior work that informed their thinking — methods they borrowed, results they built on, theories they tested against. Each of those references is a citation.

A paper with 50,000 citations has not merely been read 50,000 times. It has been actively used — woven into the fabric of 50,000 other studies. That is an entirely different kind of influence than public popularity or media coverage. It means the work became load-bearing infrastructure for a field, or in some cases, for multiple fields simultaneously.

This distinction matters because it reframes what we consider scientifically important. The most cited papers are not necessarily the most famous ones. They are the most useful ones.

Materials Science: Graphene and the Power of a Single Atom

The fifth most cited paper on the list has accumulated over 52,000 citations and concerns graphene — a material that sounds straightforward until you look closely. Graphene is a two-dimensional sheet of carbon atoms arranged in a hexagonal lattice, exactly one atom thick. Theorists in the 1940s suspected such a material might be physically impossible; something that thin should, in principle, be thermodynamically unstable.

That assumption collapsed in 2004 when researchers at the University of Manchester — Andre Geim and Konstantin Novoselov, who would later win the Nobel Prize in Physics — isolated graphene using nothing more sophisticated than adhesive tape applied repeatedly to graphite. The simplicity of the method was almost embarrassing given the magnitude of the discovery.

What followed was an avalanche of research. Graphene conducts electricity more efficiently than copper because its honeycomb lattice allows electrons to move across the surface with virtually no resistance from neighboring atoms. It is simultaneously one of the strongest materials ever tested, lighter than paper, and nearly transparent. These properties have found applications in flexible electronics, high-performance smartphone screens, advanced composites in sports equipment, and next-generation battery technology.

The review paper that became the most cited work in materials science captured this explosion of possibility at the right moment, synthesising what graphene could do and pointing toward what remained to be explored. Review papers often accumulate citations rapidly for exactly this reason: they serve as entry points for entire research communities.

Computer Science: The Deep Learning Paper Reshaping the World

If any paper on this list reflects the velocity of modern science, it is the 2015 paper "Deep Residual Learning for Image Recognition" — now standing at more than 116,000 citations. Published just a decade ago, it has already surpassed work that had been accumulating citations for generations.

To understand why, consider the problem it solved. Neural networks designed to recognise images work by passing data through successive layers of mathematical operations, each layer extracting a slightly higher-order feature — edges becoming shapes, shapes becoming objects. The deeper the network, the more sophisticated its representations. But depth creates a new problem: information degrades as it travels through dozens or hundreds of layers, like a message distorted in a long chain of telephone calls.

The paper's authors at Microsoft Research proposed a deceptively simple fix — residual connections, or "skip connections," that allow information to bypass groups of layers and feed forward directly. This means the network retains a clean signal even as it builds increasingly complex representations. The architecture became known as ResNet.

ResNet did not just improve benchmark scores. It made deep neural networks practically trainable at scales that were previously intractable, unlocking advances in medical imaging, autonomous vehicles, satellite analysis, and the computer vision components that now underpin large AI systems. Every time a system identifies a tumour in an MRI scan or a pedestrian crossing a road, there is a good chance ResNet's logic is somewhere in the chain.

Physics: Making Quantum Calculations Usable

With more than 174,000 citations, the most cited paper in physics carries the deceptively modest title "Generalized Gradient Approximation Made Simple." The title is, by the authors' own implied admission, somewhat aspirational. The subject matter — density functional theory, or DFT — is among the most technically demanding areas of quantum physics. But the paper's contribution was precisely to reduce that complexity to something workable.

DFT is the framework physicists and chemists use to calculate how electrons behave inside and around atoms. Since electrons govern how atoms bond, react, and interact, the ability to model their behaviour accurately is foundational to materials science, drug discovery, and semiconductor engineering. The problem is that electron density in a molecule is extraordinarily sensitive to the specific arrangement of atoms around it — every molecule presents a unique computational environment.

The 1996 paper by John Perdew, Kieron Burke, and Matthias Ernzerhof proposed a streamlined approximation — the PBE functional — that could describe electron density reliably using only universal physical constants, rather than parameters tuned to each individual system. This sounds like a narrow technical improvement. In practice, it was a liberation. Researchers who had previously required prohibitively complex calculations could now run DFT simulations routinely, and the PBE functional became a standard tool in computational chemistry and physics worldwide.

Its fingerprints are visible in research ranging from better photovoltaic materials to the design of pharmaceuticals to the engineering of textiles with improved breathability. The paper is a reminder that the most enduring contributions to science are often not discoveries of new phenomena, but the tools that make existing phenomena legible.

Biology: The Protein Separation Method That Never Got Replaced

Some scientific methods are simply so well-designed that they outlive every attempt to supersede them. The most cited paper in biology, published in 1970 and now exceeding 250,000 citations, introduced one such method: SDS-PAGE, short for sodium dodecyl sulphate-polyacrylamide gel electrophoresis.

The paper's formal title — "Cleavage of Structural Proteins During the Assembly of the Head of Bacteriophage T4" — gives no hint of its broader significance. What it actually did was demonstrate a reliable, reproducible way to separate proteins by molecular weight using an electric field and a gel matrix.

Here is the elegant logic: proteins come in wildly different shapes and carry varying electrical charges, which makes sorting them difficult. SDS strips away those differences. It coats every protein with a uniform negative charge proportional to its size, so when an electric field pulls them through a gel, they separate purely by mass — smaller proteins travel further, larger ones less so. The resulting banding pattern becomes a readable fingerprint of the proteins present in a sample.

Decades later, SDS-PAGE remains a daily fixture in labs diagnosing HIV, identifying cancer biomarkers, and conducting forensic analysis. The specific protocol described in that 1970 paper — run on the protein coat of a bacteriophage virus, of all things — became the standard against which every subsequent variation is compared. It is genuinely unusual for a method introduced over fifty years ago to still be considered the benchmark, but SDS-PAGE's combination of sensitivity, simplicity, and reproducibility has kept it relevant through every technological revolution since.

Chemistry: The Most Cited Scientific Paper Ever Published

At the top of the list sits a 1951 chemistry paper with more than 350,000 citations — making it, as of April 2025, the single most cited scientific publication in recorded history. The paper introduced what is now called the Lowry protein assay, a colorimetric test that measures protein concentration in a sample by producing a colour change proportional to the amount of protein present.

The underlying chemistry is simple. The assay adds copper ions to a protein sample, and these copper ions react with the peptide bonds linking amino acids together. The resulting complex then reacts with a second reagent — the Folin-Ciocalteu reagent — which turns a measurable shade of blue. More protein means more intense colour. By comparing the colour to a standard curve, a researcher can read off a concentration in minutes.

This might sound too simple to have earned 350,000 citations. But simplicity is exactly the point. The Lowry assay works across an enormous range of sample types — blood, soil, food, environmental samples — and its sensitivity is remarkably high even by modern standards. Newer assays have been developed that are faster or compatible with different reagents, but none has fully replaced the Lowry method, which consistently fills the gaps where newer technologies fail or give inconsistent results.

In a field that constantly chases novelty, the Lowry assay's longevity is a quiet form of excellence. It has been part of every major era of biological and biochemical research since the early 1950s, and there is no particular reason to expect that to change.

What the Most Cited Papers Have in Common

Look across these five papers and a pattern emerges clearly. None of them announced the discovery of a new particle, a new planet, or a new organism. None of them went viral in the popular press. What they share is something more durable: they each solved a problem that thousands of other scientists needed solved, and they solved it well enough that nobody could improve meaningfully on what they had done.

Graphene opened a material category. ResNet made deep learning trainable at scale. The PBE functional made quantum calculations accessible. SDS-PAGE gave biology a universal protein ruler. The Lowry assay gave chemistry a reliable measurement that has survived seven decades of innovation.

These are the papers that built the floor on which modern science stands. They are not always the first papers on their topics, and they are rarely the last. But they are the ones that got used — again and again, by researchers in dozens of disciplines, in thousands of labs, across generations of scientific practice.

Understanding them is not just an exercise in scientific history. It is a lesson in what actually moves knowledge forward: not always the dramatic discovery, but the precise, reliable, widely applicable tool that everyone reaches for without thinking twice.

Frequently Asked Questions

What makes a scientific paper become highly cited?

A paper accumulates citations when other researchers actively use it in their own work — referencing its methods, data, or theoretical framework. Papers that introduce practical tools, widely-used experimental protocols, or computational methods that become standard practice tend to gather the most citations over time, because they are relevant to researchers across many specialisations and time periods.

Is a high citation count the same as being scientifically important?

Not always, but there is significant overlap. Citation counts measure utility and influence within the scientific community rather than public fame or media coverage. A paper can be hugely important to practitioners — forming the methodological backbone of an entire field — while remaining completely unknown outside it. The most cited papers tend to be foundational tools rather than headline discoveries.

Why do older papers tend to have more citations?

Simply because they have had more time to accumulate them. A paper published in 1951 has had over seventy years of subsequent research to reference it. However, the rankings are sometimes disrupted by newer papers in fast-moving fields — like the 2015 ResNet paper — that accumulate citations so rapidly they surpass much older work within a decade.

Are the most cited papers in science well known to the general public?

Rarely. The most cited papers tend to describe methods, approximations, or computational techniques that are indispensable to specialists but hold little narrative interest for general audiences. Publicly famous scientific milestones — like the discovery of the structure of DNA or the detection of gravitational waves — may be culturally significant without necessarily topping citation counts, because their direct methodological utility to other researchers differs from their cultural impact.

Could an AI-generated paper ever become one of the most cited in history?

In principle, a paper containing genuinely useful methods or findings could accumulate citations regardless of how it was produced. In practice, peer review processes and editorial policies at major journals currently require meaningful human authorship and verification. The more pressing question is whether AI-assisted research tools will eventually generate the kind of broadly applicable methods — like a new assay or a new computational framework — that historically drive citation counts to extraordinary levels.