Learnt predictive behavior faces a dilemma: predictive stimuli will never 'replay' exactly as during the learning event, requiring generalization. In turn, minute differences can become meaningful, prompting discrimination. To provide a study case for an adaptive adjustment of this generalization-discrimination balance, the authors ask whether Drosophila melanogaster larvae are able to either generalize or discriminate between two odors (1-octen-3-ol and 3-octanol), depending on the task. The authors find that after discriminatively rewarding one but not the other odor, larvae show conditioned preference for the rewarded odor. On the other hand, no odor specificity is observed after nondiscriminative training, even if the test involves a choice between both odors. Thus, for this odor pair at least, discrimination training is required to confer an odor-specific memory trace. This requires that there is at least some difference in processing between the two odors already at the beginning of the training. Therefore, as a default, there is a small yet salient difference in processing between 1-octen-3-ol and 3-octanol; this difference is ignored after nondiscriminative training (generalization), whereas it is accentuated by odor-specific reinforcement (discrimination). Given that, as the authors show, both faculties are lost in anosmic Or83b(1) mutants, this indicates an adaptive adjustment of the generalization-discrimination balance in larval Drosophila, taking place downstream of Or83b-expressing sensory neurons.