The effect of training on recalling taste intensities over 6 weeks was studied using an ad libitum mixing procedure. Subjects tasted sweet and salty standards labeled as 'weak' and 'strong' (3 and 8% sucrose in redcurrant juice; 0.4 and 1.2% NaCl in beef broth). They subsequently mixed unsweetened and sweetened juice, and unsalted and salted broth, to produce taste intensities that corresponded to the standards. A minimum training (MT) group (n = 13) produced comparison stimuli by tasting and directly comparing with standards in one session only; an extensive training (ET) group (n = 13) did this in six sessions before producing comparison stimuli based on memory only at 1 h, 1 day, 1 week and 6 weeks. An upward bias (chemically determined concentrations of comparison stimuli exceeding those of standards) occurred at 1 day or 1 week in MT subjects for 'weak' and 'strong' sweetness, and for 'strong' saltiness, and sustained thereafter. The upward tendency was also observed in ET subjects but was significant only for 'strong' sweetness. It is important to recognize memory effects such as the one described, as they affect food perceptions and can be a major source of bias in sensory food research.