Controlled aging experiments were carried out in order to evaluate the changes in seed longevity occurring following the attainment of maximum seed dry weight in foxglove (Digitalis purpurea L.). Seeds harvested in three successive years were stored at ~5% moisture content (fresh weight basis) and 50°C. Seed longevity was greater the later the seeds were harvested. When the data were modelled using the predictive viability model of Ellis and Roberts, these increases were attributable to increases in both the theoretical initial viability (in probits) of the seed-lot, Ki, and in the standard deviation of the normal distribution of seed deaths in time, σ. Furthermore, there was a positive relationship between Ki and σ which was independent of the year of harvest (r = 0.9016, for 11 d.f., P < 0.001). These results do not support the predictive model which relies on the assumption that σ will not differ between seed lots of the same species stored under identical conditions. When seeds were stored at a range of moisture contents (between 5 and 10%) at 50°C, increases in σ during seed development were manifest as changes in the negative logarithmic relationship between σ and moisture content; below an upper limit, the regressions for seed lots at different stages of maturity were significantly different (P < 0.05) but could be constrained to either a common intercept or to a common slope (P < 0.05). It is proposed that it was the inherent variation in individual seed lifespans which increased during seed development. These results raise important concerns regarding the use of predetermined ‘species constants’ to predict the longevity of seed lots during long-term storage in seed banks.