In two experiments, large letters H or Z composed of small letters (also H or Z) were presented. Subjects had to make a two-choice motor response (e.g. H--left key, Z--right key). A cue presented 500 ms before the letter indicated which level (global or local) was relevant. In Experiment I, a third letter (T) sometimes appeared either at the cued or the non-cued level; in the former case, subjects had to shift their attention and to respond to stimulus features located at the non-cued level. The interference effect (RT delay in response to incongruent stimuli as compared to congruent ones) was larger when the local, rather than global, level was cued. A slow anterior negativity preceding globally-cued stimuli and shorter N1 and P2 ERP component latencies to these stimuli indicated better preparation for processing of global, as compared to local, stimulus features. The shift from local to global focus yielded a larger increase of RT, error rate, and of the P600 latency than the global-to-local shift. The P600 latency changes were parallel to those of RT. In Experiment II, the attentional shift was provoked by stimulus color red-colored letters meant that the cue was invalid, and thus, subjects had to respond to the non-cued level. Neither the interference nor the attentional shift demonstrated any asymmetry between the global and local levels. ERPs also did not differ substantially after local and global cues. In the condition demanding a shift of focus (invalid cue, incongruent letter), a positive deflection of the lateralized readiness potential indicated the activation of the wrong response channel. The large RT increment in this condition was not accompanied by an increase of the P600 latency. Two possible mechanisms of attentional shift may be proposed, the first related to perceptual processes (e.g. an additional visual search), and the second, to the competition between two response intentions.