This paper applied Donders' subtraction method to examine the processing of global and selective stop signals in the stop-signal paradigm. Participants performed on three different versions of the stop task: a global task and two selective tasks. A global task required participants to inhibit their response to a go signal whenever a stop signal was presented (Stop-a task). A selective stop task required participants to inhibit to one stop signal but not to the other (Stop-c task). Another selective stop task required them to inhibit when the response indicated by go and stop signals was the same but not when they were different (Stop-b task). Stop-signal reaction time (SSRT) was shortest for Stop-a and longest for Stop-b, with intermediate values for the Stop-c task. Additional control experiments that manipulated stop probability confirmed the robustness of global and selective stopping latencies even when the stop-signal probability varied. The current findings contribute to the conclusion that Donders' subtraction method provides a useful tool for estimating the durations of subprocesses that together comprise SSRT.