An evolutionary landscape is a metaphor or a construct used to think about and visualize the processes of evolution (e.g. natural selection and genetic drift) acting on a biological entity (e.g. a gene, protein, population, or species). This entity can be viewed as searching or moving through a search space. For example, the search space of a gene would be all possible nucleotide sequences. The search space is only part of an evolutionary landscape. The final component is the "y-axis", which is usually fitness. Each value along the search space can result in a high or low fitness for the entity. If small movements through search space cause changes in fitness that are relatively small, then the landscape is considered smooth. Smooth landscapes happen when most fixed mutations have little to no effect on fitness, which is what one would expect with the neutral theory of molecular evolution. In contrast, if small movements result in large changes in fitness, then the landscape is said to be rugged. In either case, movement tends to be toward areas of higher fitness, though usually not the global optima.

What exactly constitutes an "evolutionary landscape" is frequently confused in the literature; the term is often used interchangeably with "adaptive landscape" and "fitness landscape", although some authors have different definitions of adaptive and fitness landscapes. Additionally, there is a large disagreement whether the concept of an evolutionary landscape should be used as a visual metaphor disconnected from the underlying math, a tool for evaluating models of evolution, or a model in and of itself used to generate hypotheses and predictions.

According to McCoy (1979), the first evolutionary landscape was presented by Armand Janet of Toulon, France, in 1895. In Janet's evolutionary landscape, a species is represented as a point or an area on a polydimensional surface of phenotypes, which is reduced to two dimensions for simplicity. The size of the population is proportional to the amount of variation within the population. Natural selection (the influence of the exterior features) is represented by a vector. Unlike the evolutionary landscapes of those who would follow, in Janet's concept, natural selection pulls species toward the minima instead of the maxima. This is because the y-axis does not represent fitness but stability. One important aspect of Janet's evolutionary landscape (versus Wright's) is that the landscape changes as the environment changes.

Credit for the first evolutionary landscape typically goes to Sewall Wright, and his idea has arguably had a much larger audience and greater influence on the field of evolutionary biology than any other comparable understanding of "evolutionary landscape". In his 1932 paper, Wright presents the concept of an evolutionary landscape composed of a polydimensional array of gene (allele) or genotype frequencies and an axis of fitness, which served as a visual metaphor to explain his shifting balance theory. Similarly to Janet, Wright felt the landscape could be reduced to two dimensions for simplicity. (This is one of the greatest criticisms, which are discussed below.) Populations are represented by areas, with the size of the area corresponding to the amount of genetic diversity within the population. Natural selection drives populations toward maxima, while drift represents wandering and could potentially cause a peak shift. Movement across the landscape represented changes in gene frequencies. This landscape was represented as a series of contour lines, much like a topographical map; while selection kept or moved a biological entity to a peak, genetic drift allowed different peaks to be explored.

In 1944, Simpson expanded Wright's landscape to include phenotypes. In Simpson's model, the landscape is a means of visualizing the "relationship between selection, structure, and adaptation." Unlike Wright, Simpson used the landscape to represent both natural selection and genetic drift. Uphill movements are due to positive selection, and downhill movements are due to negative selection. The size and shape of a peak indicated the relative specificity of selection; i.e. a sharp and high peak indicates highly specific selection. Another difference between Simpson's and Wright's landscapes is the level at which evolution is acting. For Wright, a population geneticist, only populations of a species were shown. In Simpson's figures, the circles drawn represent all of Equidae. The most important difference is that in Simpson's model, the landscape could vary through time, while in Wright's model, the landscape was static. Wright reviewed Simpson's work (Tempo and Mode in Evolution) and did not object to Simpson's use of evolutionary landscapes. In later writings, Simpson referred to peaks as adaptive zones.

In a series of papers, Russell Lande developed a mathematical model for Simpson's phenotypic landscape. Lande reconciled Wright's population-level view with Simpson's use of higher taxonomic levels. Lande considers fitness peaks to be determined by the environment and thus represent ecological niches or adaptive zones for a population. Clusters of peaks inhabited by phenotypically similar populations can be viewed as higher taxonomic levels.

The concept of evolutionary landscapes changed once more as the modern understanding of molecular evolution emerged. It is claimed that Maynard Smith (1970) was the first to visualize protein evolution as a network of proteins one mutational step away from others. However, for this to be true, there must be pathways between functional proteins. Acknowledging the work of Kimura, King, and Jukes (the neutral theory of molecular evolution), Maynard Smith realized the proteins along such pathways could have equal functionality or be neutral. In other words, not all moves in evolution are "uphill". In 1984, Gillespie adapted the concept of evolutionary landscapes to nucleotide sequences and so visualized the "mutational landscape" whereby all nucleotide sequences are one mutational step away from another, which is remarkably similar and yet fundamentally different from Wright's original concept. This conceptual shift, along with the development of vast computational power, has allowed evolutionary landscapes to move from being a simple visual metaphor to a working model of evolution. As one might expect, this has drawn heavy criticism and generated much research.

One of the first criticisms (or at least difficulty) with evolutionary landscapes is their dimensionality. Wright recognized that true landscapes can have thousands of dimensions, but he also felt reducing those dimensions to two was acceptable since his point in doing so was simply to convey a complex idea. As a visual metaphor, this might be a valid reduction; however, the work of Gavrilets has shown that taking the high dimensionality of evolutionary landscapes into consideration may matter. In a high-dimensional framework, the peaks and valleys disappear and are replaced with hypervolume areas of high fitness and low fitness, which can be visualized as curved surfaces and holes in a three-dimensional landscape. While this does not affect visualization of the landscape per se (i.e. holes are equivalent to valleys), it does affect the underlying mathematical model and the predicted outcomes.