Inefficient visual search can become efficient with practice [Vision Research 35 (1995) 2037; 40 (2000) 2925]. In this study, we wondered whether this improvement depends on unique visual features associated with the target, on differences in item-specific brightness distribution between target and distractors, or only on a change in the allocation of attention and thus global search strategy. We found that both, unique visual features and differences in brightness distribution lead to parallelisation with practice of originally inefficient search. Prolonged practice of inefficient search tasks lacking both unique visual features and differences in brightness distribution (conjunctions) does not lead to improved performance, thus indicating that perceptual learning in visual search does not solely reflect an unspecific global improvement in search strategy. Changing the brightness polarity of the stimuli leads to instantaneous, complete transfer to the new task. There is no transfer but rather trade-off between the learning based on unique visual features or on differences in brightness distribution between target and distractors.