Here we present results from three experiments demonstrating that the way in which options are organized during learning (i.e., the choice architecture) significantly affects the resulting memory representations of their economic values. That is, options that are optimal in the learning contexts tend to be significantly overvalued in the follow-up memory tests. By changing the choice architecture of the learning phase across experiments, we were able to show that this irrational bias is a direct consequence of the learning choice architecture, since presenting options in all possible combinations during learning eradicates this effect. Critically, all the results stand irrespective of the memory elicitation used.