Pulses of transcranial magnetic stimulation (TMS) with a predominantly anterior-posterior (AP) or posterior-anterior (PA) current direction over the primary motor cortex appear to activate distinct excitatory inputs to corticospinal neurons. In contrast, very few reports have examined whether the inhibitory neurons responsible for short-interval intracortical inhibition (SICI) are sensitive to TMS current direction. To investigate whether SICI evaluated with AP and PA conditioning stimuli (CSPA and CSAP) activate different inhibitory pathways. SICI was always assessed using a PA-oriented test stimulus (TSPA). Using two superimposed TMS coils, CSPA and CSAP were applied at interstimulus intervals (ISI) of 1-5 ms before a TSPA, and at a range of different intensities. Using a triple stimulation design, we then tested whether SICI at ISI of 3 ms using opposite directions of CS (SICICSPA3 and SICICSAP3) interacted differently with three other forms of inhibition, including SICI at ISI of 2 ms (SICICSPA2), cerebellum-motor cortex inhibition (CBI 5 ms) and short-latency afferent inhibition (SAI 22 ms). Finally, we compared the effect of tonic and phasic voluntary contraction on SICICSPA3 and SICICSAP3. CSAP produced little SICI at ISIs = 1 and 2 ms. However, at ISI = 3 ms, both CSAP and CSPA were equally effective at the same percent of maximum stimulator output. Despite this apparent similarity, combining SICICSPA3 or SICICSAP3 with other forms of inhibition led to quite different results: SICICSPA3 interacted in complex ways with CBI, SAI and SICICSPA2, whereas the effect of SICICSAP3 appeared to be quite independent of them. Although SICICSPA and SICICSAP were both reduced by the same amount during voluntary tonic contraction compared with rest, in a simple reaction time task SICICSAP was disinhibited much earlier following the imperative signal than SICICSPA. SICICSPA appears to activate a different inhibitory pathway to that activated by SICICSAP. The difference is behaviourally relevant since the pathways are controlled differently during volitional contraction. The results may explain some previous pathological data and open the possibility of testing whether these pathways are differentially recruited in a range of tasks.