The growth–mortality relationship was assessed for larval cohorts of the Japanese sardine Sardinops melanostictus using two new approaches: (1) repeat measures in general linear model (RM-GLMs) and (2) the autoregressive-individual method (AIM). Both methods were compared to the traditional approach in which repeat-measure ANOVA was used to compare the changes in increment width (WI) at age and otolith radii (RO) at age between individuals from an original population and survivors. In RM-GLMs, both the WI at age and RO at age (i.e. at 5, 10, 15 and 20 days) were used as the dependent variables, and the standardized residuals of both regressions RO and age and RO and total length (LT), age class, and day of the year as independent variables. A significant increase in WI at age and RO at age from younger to older age classes was seen as indicative of growth-dependent selection. In AIM, the RO-at-age relationship for each fish was fitted for the first 20 days, using autoregression, and then the growth traits (i.e. slopes) between the original cohorts and survivors were compared using ANOVA. In the traditional approach, the WI at age and RO at age for the first 20 days of an original population were compared with those of survivors sampled in later stages. GLMs and traditional approaches supported the growth rate (i.e. the faster an individual grows, the higher its probability of survival) and bigger is better (i.e. larger individuals at any given age will have lower probability of mortality than smaller individuals of the same age) mechanisms. Furthermore, AIM showed that individuals from original cohorts had lower otolith growth rates than those from survivors, giving further support for the growth–mortality hypothesis for the larval stage of this clupeid.