The motor training hypothesis, proposed in its first form nearly half a century ago and broadened subsequently, states that the function of play is adaptive modification of the developing neuromuscular system. Evidence from many mammalian species indirectly supports the motor training hypothesis, but the exact nature of developmental change prompted by play remains unknown. We reviewed literature on the anatomical and physiological effects of exercise in mammals and categorized these as effects available to individuals at any age, versus effects available only during a discrete period of postnatal development, and transitory effects, which decay soon after exercise ends, versus permanent effects. We found that most effects are available at any age and are transitory; we argue that they are not likely primary benefits of play. However, two effects that influence motor performance-modification of cerebellar synaptogenesis and modification of skeletal muscle fiber type differentiation-are available only during a short period of postnatal development and appear to be permanent. In three species for which both kinds of data were available, the age distribution of play closely matched the age distribution of these two types of experience-modifiable development. We propose that play may not be motor training in the broad sense, but rather it may be behavior designed to influence specific types of development.