||This paper summarizes what is known about farm-level maize seed management practices and reviews the theoretical and empirical evidence regarding the relationship between farmers’ seed recycling practices and the genetic composition (and agronomic performance) of maize cultivars. The focus is on farmers in developing countries, many of whom do not replace their seed annually with newly purchased commercial seed but rely instead on recycled seed saved from their own harvest or obtained from other farmers. Why is it important to know about the genetic composition of maize plants found in farmers’ fields? Although there are many possible reasons, for research organizations such as CIMMYT that carry out plant breeding activities one of the most important is to be able to calculate the value of improved germplasm. Modern varieties of maize (MVs) have been a major source of productivity growth in the past and are likely to be and increasingly important source in the future. In order to calculate the economic value of MVs (which is needed to determine the optimal level of investment in maize breeding research), it is necessary to estimate the productivity gains associated with adoption of improved germplasm. These productivity gains cannot be estimated unless it is possible to identify unequivocally the materials growing in farmers’ fields. Many empirical studies make clear that maize farmers in developing countries frequently save seed from their own production to replant the following season. By far the most common seed selection practice is post-harvest selection. Although there are a number of obvious advantages associated with selecting kernels from harvested ears, the practice does not always result in the production of genetically pure seed. Largely for this reason, recycling is often associated with changes in the genetic composition of maize cultivars. What happens, genetically speaking, when farmers save maize seed from their own harvest and replant it the following cropping cycle? Based on what is known about the reproductive biology of maize, as well as farmers’ varietal management practices and seed selection strategies, there are strong reasons to expect that the genetic composition of farmer-maintained cultivars will change over time. Seven potential sources of genetic change in recycled maize can be distinguished: (1) farmers’ seed selection practices, (2) unintentional seed mixing, (3) contamination, (4) genetic drift, (5) mutation, (6) natural selection, and (7) segregation. Each of these is discussed, and published studies are reviewed to determine whether theoretical predictions about the amount of genetic change attributable to each source are supported by empirical evidence. Our review of the literature suggests that landraces, improved open-pollinated varieties (OPVs) and hybrids all undergo changes in genetic composition as a result of seed recycling. The sources of these changes vary in importance by type of material. In landraces and improved OPVs, genetic changes result from a combination of intentional and unintentional selection pressure. Landraces and OPVs evolve not only because farmers deliberately select for desired characteristics, but also because of environmental influences, iv accidental cross-pollination, random mutation, and gene segregation. Since both types of selection pressure are highly variable, it is difficult to generalize about the rate of genetic change; depending on the circumstances, the genotype of a landrace or improved OPV can change significantly from one generation of plants to the next, or it can remain essentially unchanged across many generations of plants. In hybrids, by far the most important source of genetic changes is segregation — random recombination of alleles that occurs when seed is recycled. Key results of a simulation exercise designed to show the likely effects of inbreeding in maize hybrids appear to be supported by findings published in the empirical literature on seed recycling: When maize hybrids are recycled, yield usually decreases significantly from the F1 to the F2 generation. Yield tends to stabilize in subsequent generations, however, and may eventually begin to increase again if farmers are exerting selection pressure. When maize hybrids are recycled, the size of the yield decrease observed between the F1 and F2 generations depends in large part on the level of inbreeding of the original parents. Generally speaking, the greater the degree of inbreeding in the parents, the greater the degree of heterosis in the F1 generation, and the greater the yield decline observed between the F1 and F2 generations. This relationship may be confounded by environmental factors, however. The degree of inbreeding of the parents affects not only the size of the expected yield decrease but also its variability. The greater the level of heterozygosity in F1 plants, the greater the variability in inbreeding depression expected in F2 and F3 plants. Whether or not advanced-generation hybrids outyield landraces and improved OPVs depends on the original difference in yield and on the magnitude of the decline in yield caused by recycling. In some instances, recycled hybrids continue to outyield the other types of materials, which explains why hybrid recycling may make sense. Recycling of hybrids may have little effect on qualitative traits such as kernel size and shape, grain texture, and pounding quality. The finding that seed recycling often leads to significant genetic changes in farmermaintained cultivars suggests that there may be a need to reassess the categories traditionally used to classify maize varieties (e.g., landraces, improved OPVs, hybrids). In addition, the rapid rate of genetic change observed to take place in farmers’ fields has important implications for research impacts assessment studies. Practical guidelines for use in estimating the returns to maize breeding research are presented in the appendix.