The two-phase sampling design has been traditionally used in ecology applications to estimate age composition. With regard to both economic and statistical considerations, however, the two-phase design may yield estimates of the age composition that are no better than can be obtained using simple random sampling. This article shows that the optimality of the two-phase sampling design depends upon (a) survey objectives, (b) per-unit sampling costs, and (c) the predictive power of stratum membership on age. Methods are given for determining whether two-phase sampling yields better estimates of the age composition than simple random sampling, accounting for (a), (b), and (c).