Experiment 1 During the summer of 2005, seeds collected in 1999 from 30 maternal families per population were germinated. Two weeks after germination, 30 plants per population (one plant per family × 30 families × four populations, n = 120) were individually transplanted into 4-l pots, filled with potting soil, and placed in a glasshouse at the Instituto de Ecología (UNAM). Simultaneously, c. 100 adults of L. trilineata from each population were collected and taken to the laboratory. Herbivores were allowed to reproduce for 1 wk to obtain c. 30 clutches per population (n = 827). After eclosion (August 2005), each clutch was divided in four groups so that each group could be fed with leaves from each plant population. There were at least six larvae per clutch per plant population. All the larvae were individually reared and their survival was checked daily until adult emergence (for a description of the rearing technique see Espinosa & Fornoni, 2006). Mean survival time was calculated for each plant–herbivore combination. For each individual larva, the efficiency of food consumption, developmental time and adult mass was also recorded. Efficiency was estimated as weight gained relative to the proportion of leaf area consumed. Weight gain was estimated as the increment in mass between the second and fourth larval stages relative to the number of days between these two stages. Leaf area consumed by each larva was calculated using a Digital Image Analysis System (WinDias Basic; Delta-T Devices Ltd, Cambridge, UK). Developmental time was considered as the number of days between hatching and adult emergence. Adults were weighted to the nearest mg using a digital balance (OHAUS, Parsippany, NJ, USA). To obtain a more accurate fitness estimate, individual performance was calculated as the product of the relativized values of mean survival time, efficiency of food consumption, inverse of developmental time and adult mass. Therefore, high values of individual performance should be interpreted as larvae that, on average, survived more days, were more efficient, completed their development quickly and were bigger as adults. For all these variables, a significant interaction between herbivore and plant population was detected previously, supporting the assumption that all these variables could be responsible for the variation in performance (see the Supporting Information, Table S1). Because herbivore performance (Table S2) was estimated as the product of four variables, it is possible that high correlations between these variables could overestimate the values of performance. In order to estimate the severity of multicollinearity, the variance inflation factors (VIF) were calculated for each variable (Neter et al., 1996; p. 386). All VIFs were between 1.03 and 1.08, which means that there is no overestimation of the F ratio in the ANOVA for herbivore performance.