Theories for optimal life history strategies in variable environments have until now focused on cases where the individuals have either no information about the environment (models maximizing geometric mean fitness) or full information about the environment (models predicting optimal reaction norms). In this paper I investigate the optimal time for multivoltine organisms to commence seasonal reproduction in a more general and realistic case where animals perceive the state of their environment through cues that are measured with varying degrees of precision. If there were only a trade-off between early reproduction and high reproductive success, and if animals had perfect information about their environment, it would be optimal to commence reproduction when the rate of change in reproductive success relative to its current value equals the difference between population growth during the reproductive and non-reproductive seasons. This implies that reproductive success at the optimum is independent of when (but not how) the environment improves over the season. However, because it is optimal to respond conservatively to uncertain cues, we should expect higher reproductive success during years when breeding conditions improve early than when they improve late. Nevertheless, a phenotypic correlation between reproductive success and timing of reproduction will probably not be detectable in a stochastic environment. Data from a cyclic population of field voles (Microtus agrestis L.) in northern England show a negative correlation between reproductive success and timing of reproduction among out-of-phase locations. Such a pattern may occur when there is a convex trade-off between pre-breeding survival and timing of reproduction, or if animals precipitate reproduction to avoid senescence when the environment improves late.