In the lack of a satisfactory formulation for the two-stanza growth curve of yellowfin (Thunnus albacares), estimates of growth rate by size class can be an alternative as an input for integrated models. From tagging data collected in the Atlantic and Indian Oceans, we show however that for growth rate that does not follow a classical linear decreasing trend over size, simple estimates of length increments by size are biased due to an averaging effect of the time spent at liberty. We simulate length at recapture to illustrate the magnitude of the bias which can be expected depending on the duration of the time at liberty. Our results suggest that for large time at liberty growth rates by size are smoothed in such a way that the Von Bertalanffy model could erroneously be considered as a good alternative. With the aim of removing this bias we used a generalized additive model (GAM) to calculate an instantaneous growth rate by size (i.e., fixing time at liberty at 1 day at sea). The predicted growth rates by size are lower than the apparent growth rates but are similar in magnitude and in shape with observed growth rates for moderate times at liberty (< 90 days at sea). In the light of our findings, we suggest that hard part reading analyses, which in general reject the two-stanza growth assumption, could have been affected by the same type of bias.