Sampling protocols for pest management programs often rely on phenology models to predict the time of occurrence of the stage to be sampled. In some cases it may not be possible or practical to predict occurrence with sufficient accuracy to ensure the reliability of a sampling protocol based on a single sampling session. We describe a sequential classification procedure that is useful for such situations. The method compensates for uncertainty in phenological prediction by allowing for sampling on more than one date. Each sampling session uses a sequential sampling plan with thresholds based on a range of possible values of cumulative recruitment at the time of sampling. These values are given by a set of phenological models derived from field data that describe recruitment over time. Because the sampling sessions are linked in time, we call the procedure cascaded sequential sampling . We describe development of the cascaded sequential sampling plan using a simple hypothetical case, and demonstrate its application to the problem of sampling second-generation spotted tentiform leafminer, Phyllonorycter blancardella (F.), in western New York apples.