The forest insect pest Bupalus piniarius (pine looper moth) is a classic example of a natural population cycle. As is typical for Populations that exhibit regular oscillations in density, there are several biological mechanisms that are hypothesized to be responsible for the cycles; but despite several decades of detailed study there has been no definite conclusion as to which mechanism is most important. We evaluated three hypotheses for which there was direct experimental evidence: (1) food quality (nutritional value of pine needles affected by defoliation); (2) parasitoids (trophic interactions with specialist parasitoids), and (3) maternal effects (maternal body size affects the performance of offspring). We reviewed the empirical evidence for each of these hypotheses and expressed each hypothesis in the form of a mechanistic dynamic model. We used a nonlinear forecasting approach to fit each model to three long-term Population time series in Britain that exhibit some degree of regular cycling, and we used parametric bootstrap to evaluate the significance of differences between models in their goodness of fit to the data. The results differed among the three forests: at Culbin, the parasitoid and maternal effects models fit equally well; at Roseisle, the food quality and maternal effects models fit equally well; and at Tentsmuir, the parasitoid model fit best. However, the best-fit parasitism models required that the parasitism rate vary between nearly 0 and nearly 1 during a cycle, greatly exceeding the range of parasitism rates that have been observed in the field. In contrast, the required variation in the observable maternal quality variable (pupal mass) was within the range of empirical observations. Under mild constraints on the parasitism rate (though allowing a much wider range than has been measured in B. piniarius at any location), the fit of the parasitism model fell off dramatically. The maternal effects model then had uniformly strong support, outperforming the constrained parasitism model at all three sites and the food quality model at two; it performed slightly better than the food quality model at the remaining site. This represents the first system in which the maternal effects hypothesis for population cycles has been supported by both strong biological and dynamical evidence.