Inhibition of motor responses is a concealed operation. It becomes manifest only through the absence of expected behavior. Therefore, some widely applicable methods and dependent variables from experimental psychology are inapplicable for inhibition research. This problem applies to performance (e.g., the absence of a reaction time) as well as psychophysiological reflections of inhibition (e.g., uncertainty about the moment of inhibition as a marker). We believe that this problem has led to paradoxical answers to the question where responses are inhibited, and that this is partly due to indistinctness about the exact question. In the literature, different meanings of the concept ‘locus of inhibition’ are intermingled, thus giving rise to confusion about the agent, site, and manifestation of inhibition. This paper critically reviews data that have been put forward in support of loci in subcortical, cortical, and peripheral areas of the nervous system.

Inhibition needs to be investigated through a comparison between conditions with and without response execution. It can be investigated in a laboratory setting with the stop-signal paradigm and the go/no-go paradigm. Although there may be other paradigms that probe the same inhibitory mechanisms, we only discuss inhibition of the type that is required on these two paradigms. On a typical stop-signal task, a subset of trials from a series of regular choice reaction time (RT) trials is interrupted by a stop signal. The stop signal instructs the subject to withhold the response that was in preparation if possible. It becomes harder to suppress a response as the stop signal is presented closer to the moment of responding. A profile of inhibitory efficiency over time is derived by manipulating the stimulus onset asynchrony (SOA) between the response and stop signal. The stop-signal paradigm can thus be regarded as an elaboration of the go/no-go task, because in the go/no-go task, the SOA is always zero. The stimulus configuration often combines inhibition and response signals, so that the instruction can be, for example, to respond to a character (go), but not to respond when the character is presented in red (no-go).

Several authors have suggested that, in order to be successful, inhibitory processes have to win a race against concurrent response processes. If inhibitory processes finish before the response processes do, the response is correctly withheld; otherwise, the response is executed. The formal version of the ‘horse-race model’ gives a rather powerful description of inhibitory control, because it is able to account mathematically for performance in the stop-signal paradigm and allows the calculation of inhibitory speed. It assumes that response processes proceed independent of stop processes.

For the calculation of stop speed, it is only necessary to know the latencies of starting and finishing the stop processes. The difference between these two latencies is known as the stop-signal reaction time (SSRT). The SOA defines the start of the stop processes, and the finish needs to be derived from other results. Given the independence assumption, the distribution of RTs on trials without a stop signal (nonsignal trials) can be used as an approximation of the covert distribution of signal RTs. The latter distribution is not entirely available because it is trimmed on the right-hand side by inhibition. It is known how much of the distribution is trimmed, because this part equals the proportion of inhibition trials. Consequently, the boundary that distinguishes between inhibition on the right and execution on the left side of the nonsignal RT distribution can be interpreted as the average finish of the stop processes.

The SSRT, which can be conceived of as an index of inhibitory efficiency, does not vary much with the primary task. It varies between 200 and 250 ms for normal adults, and is somewhat prolonged for children and older adults. In addition, there have been reports of group differences in SSRT related to impulsivity in normal subjects and children with attention deficit hyperactivity disorder.

Variations of the stop-signal paradigm have addressed complications of the stop processes, and the effect of the primary task onto stop processes. For this review, an interesting complication of stop processes is the requirement to stop a response, while executing an alternative response. In this stop-change condition, the inhibition latency is found to be longer than in the stop-all condition – the condition that is created by a regular stop instruction. This difference in speed could simply reflect the refinement of the requirement to stop. It has been suggested, however, that there may be two separate mechanisms for stop-all and stop-change conditions.

Logan and Burkell (1986) performed an experiment with a two-and four-choice primary task. In two stop-selective conditions, the stop signal applied only to one response, and not to the one or three other responses. SSRT was longer for selective stopping in the four-choice than in the two-choice task, and this was not the case in the stop-all conditions. Logan et al. (1997) suggested that there is a global mode for nonselective and a local mode for selective inhibition. De Jong et al. (1995), however, concluded that inhibition on a stop-selective condition requires the same mechanism as stop-all inhibition. Note that the suggestion of two modes does not imply separate mechanisms. Both for the stop-change and the stop-selective condition, it can be argued that the duration of stopping increases with the refinement of control, following a stop-signal equivalent of Hick’s law.