Focusing of attention to a specific speech source plays an important role in everyday speech perception. However, little is known of the neuronal substrates of focused attention in speech perception. Thus, the present study investigated the effects on neuronal activation of directed attention to auditory stimuli that differed in semantic content. Using an event-related fMRI protocol, single vowels, three-phoneme pseudowords and three- and four-phoneme real nouns and words were randomly presented to the subjects during four different instructional conditions. One condition was passive listening without any specific instructions of focusing of attention. The other conditions were attention focused on either the vowels, the pseudowords or the words. Thus, the acoustic stimulation was constant across conditions. The subjects were 13 healthy adults. Functional MRI was performed with a 1.5 T scanner, using an event-related design. During passive listening, there were significant activations bilaterally in the superior temporal gyrus. Instruction to attend to the pseudowords caused activation in middle temporal lobe areas, extending more anterior compared to the activations seen during passive listening. Instruction to attend to the vowel sounds caused an increase in activation in the superior/medial temporal lobe, with a leftward asymmetry. Instruction to attend to the words caused a leftward asymmetry, particularly in the middle and superior temporal gyri. It is concluded that attention plays a modulatory role in neuronal activation to speech sounds, producing specific activations to specific stimulus categories that may act to facilitate speech perception.