Hard science and soft science are colloquial terms used to compare scientific fields on the basis of perceived methodological rigor, exactitude, and objectivity. In general, the formal sciences and natural sciences are considered hard science by their practitioners, whereas the social sciences and other sciences are described by them as soft science.

Precise definitions vary, but features often cited as characteristic of hard science include producing testable predictions, performing controlled experiments, relying on quantifiable data and mathematical models, a high degree of accuracy and objectivity, higher levels of consensus, faster progression of the field, greater explanatory success, cumulativeness, replicability, and generally applying a purer form of the scientific method.^{[excessive citations]} A closely related idea (originating in the nineteenth century with Auguste Comte) is that scientific disciplines can be arranged into a hierarchy of hard to soft on the basis of factors such as rigor, "development", and whether they are basic or applied.

Philosophers and historians of science have questioned the relationship between these characteristics and perceived hardness or softness. The more "developed" hard sciences do not necessarily have a greater degree of consensus or selectivity in accepting new results. Commonly cited methodological differences are also not a reliable indicator. For example, social sciences such as psychology and sociology use mathematical models extensively, but are usually considered soft sciences. While scientific controls are cited as a methodological difference between hard and soft sciences, in certain natural sciences, like astronomy and geology, it is impossible to perform controlled experiments to test most hypotheses and observation and natural experiments are primarily used instead. Survey data about the replication crisis among researchers strongly suggests that the failure to reproduce published findings has impacted the natural and applied sciences along with psychology and the social sciences. However, there are some observable differences between hard and soft sciences. For example, hard sciences make more extensive use of graphs, and soft sciences are more prone to a rapid turnover of buzzwords.

The metaphor has been criticised for unduly stigmatizing soft sciences, creating an unwarranted imbalance in the public perception, funding, and recognition of different fields.

The origin of the terms "hard science" and "soft science" is obscure. The earliest attested use of "hard science" is found in an 1858 issue of the Journal of the Society of Arts, but the idea of a hierarchy of the sciences can be found earlier, in the work of the French philosopher Auguste Comte (1798‒1857). He identified astronomy as the most general science, followed by physics, chemistry, biology, then sociology. This view was highly influential, and was intended to classify fields based on their degree of intellectual development and the complexity of their subject matter.

The modern distinction between hard and soft science is often attributed to a 1964 article published in Science by John R. Platt. He explored why he considered some scientific fields to be more productive than others, though he did not actually use the terms themselves. In 1967, sociologist of science Norman W. Storer specifically distinguished between the natural sciences as hard and the social sciences as soft. He defined hardness in terms of the degree to which a field uses mathematics and described a trend of scientific fields increasing in hardness over time, identifying features of increased hardness as including better integration and organization of knowledge, an improved ability to detect errors, and an increase in the difficulty of learning the subject.

In the 1970s sociologist Stephen Cole conducted a number of empirical studies attempting to find evidence for a hierarchy of scientific disciplines, and was unable to find significant differences in terms of core of knowledge, degree of codification, or research material. Differences that he did find evidence for included a tendency for textbooks in soft sciences to rely on more recent work, while the material in textbooks from the hard sciences was more consistent over time. After he published in 1983, it has been suggested that Cole might have missed some relationships in the data because he studied individual measurements, without accounting for the way multiple measurements could trend in the same direction, and because not all the criteria that could indicate a discipline's scientific status were analysed.

In 1984, Cleveland performed a survey of 57 journals and found that natural science journals used many more graphs than journals in mathematics or social science, and that social science journals often presented large amounts of observational data in the absence of graphs. The amount of page area used for graphs ranged from 0% to 31%, and the variation was primarily due to the number of graphs included rather than their sizes. Further analyses by Smith in 2000, based on samples of graphs from journals in seven major scientific disciplines, found that the amount of graph usage correlated "almost perfectly" with hardness (r=0.97). They also suggested that the hierarchy applies to individual fields, and demonstrated the same result using ten subfields of psychology (r=0.93).