The main aim of the presented study was to search on specific stimulation conditions that determine the process of attentional shifting. The selective attention task TPSA based on Posner and Mitchell's paradigm was applied (N = 71). In compliance with the task instructions, subjects had to make an elastic change of the selection rule (nominal identity, e.g. <r,R>/physical identity, e.g. <R,R>). The main experimental manipulations were regularity of the task switches (predictable/unpredictable) and duration of the breaks between tasks (0-400-600-800 ms ). The obtained effect of break duration reflected automatic costs of attentional shifting. However, the predictability of selection rules did not improve the attentional performance. It has been also shown that only switching to deeper level of information processing costs a lot, whereas switching to shallow levels is quite effortless. The data is analyzed with regard to Johnston model of flexible filter of selective attention.