In BOLD fMRI a detailed analysis of the MRI signal time course sometimes shows time differences between different activated regions. Some researchers have suggested that these latencies could be used to infer the temporal order of activation of these cortical regions. Several effects must be considered, however, before interpreting these latencies. The effect of a slice-dependent time shift (SDTS) with multi-slice acquisitions, for instance, may be important for regions located on different slices. After correction for this SDTS effect the time dispersion between activated regions is significantly decreased and the correlation between the MRI signal time course and the stimulation paradigm is improved. Another effect to consider is the latency which may exist between perception and stimulus presentation. It is shown that the control of perception can be achieved using a finger-spanning technique during the fMRI acquisition. The use of this perception profile rather than an arbitrary waveform derived from the paradigm proves to be a powerful alternative to fMRI data processing, especially with chemical senses studies, when return to baseline is not always correlated to stimulus suppression. This approach should also be relevant to other kinds of stimulation tasks, as a realistic way of monitoring the actual task performance, which may depend on attention, adaptation, fatigue or even variability of stimulus presentation.