An ongoing question in functional MRI is precisely how measured signal changes relate to neuronal activity. While this question has been probed using animal models and electrophysiologic measures of neuronal activity, it has also been probed by examining, in humans, the spatial location, magnitude, and temporal dynamics of signal changes to well understood stimuli. With regard to dynamics, several earlier studies have revealed a larger than expected response to brief stimuli, hypothesized to result from nonlinearities in either the hemodynamics or the neuronal activity. In this study, we investigate the linearity of the increase in blood oxygenation level dependent (BOLD) contrast as a function of stimulus duty cycle, as well as the linearity of the decrease in BOLD as a function stimulus "off" duration. These findings not only shed further light on the mechanisms behind BOLD contrast but also give practical information as to what to keep in mind when performing and interpreting event related fMRI experiments. These experiments demonstrated: a) the BOLD signal decrease, on stimulus cessation, was smaller than predicted by a linear system--opposite to what has been reported in the literature associated with a signal increase, and b) the deconvolved event-related BOLD signal is highly dependent on duty cycle (the fraction of time activated vs. non-activated), Several potential mechanisms explaining these dynamics are discussed and modeled. We find that the experimental results are most consistent with a nonlinear neuronal response, but do not rule out significant effects of nonlinear hemodynamic factors, in particular the nonlinear relationship between oxygen extraction fraction and blood flow.