The choice of food plants often assumes critical importance for a herbivore. Although many studies have investigated host‐plant choice behavior, few have examined preferences (vs. growth and survival) at multiple stages of the life cycle, notwithstanding the importance of identifying the critical stage(s) in an animal’s life history. Fern moths Herpetogramma theseusalis (Lepidoptera: Crambidae) provide an excellent opportunity to test host‐plant choice at several stages. Fern moth larvae feed on distantly related ferns, sensitive Onoclea sensibilis and marsh fern Thelypteris palustris, and adults oviposit on both species. We examined newly hatched larvae, overwintered larvae and ovipositing females to test hypotheses predicting when host‐plant choice takes place (overwintering and mobility hypotheses: overwintering stage determines choice of substrate vs. most mobile stage chooses) and the basis for choice (optimal oviposition and enemy‐free space hypotheses: resource producing highest fecundity vs. lowest losses to enemies). We also evaluated the hypothesis that host‐associated fitness trade‐offs explain host specialization. Only ovipositing females, the most mobile stage, exhibited a clear preference (for marsh fern), consistent with the mobility hypothesis. However, their preference for marsh fern fits neither the optimal oviposition hypothesis nor the enemy‐free space hypothesis; although some larvae initially grew faster on marsh fern, adults reared from the two ferns did not differ significantly in mass and experienced marginally lower parasitism on sensitive fern. Thus, we found no host‐associated fitness trade‐offs. Overwintering losses in marsh fern plots exceeded those in sensitive fern, and mixed plots supported the most overwintered larvae. Preference for marsh fern suggests that early success drives host‐plant choice, an advantage that later disappears. Temporal variability may prevent closer fits to the hypotheses, because both ferns provide the moths with acceptable resources throughout their life cycles.