Objective Recently, language mapping by repetitive transcranial magnetic stimulation (rTMS) has gained a lot of interest in preoperative planning to preserve language function. However, the improvement of rTMS protocols is still a matter of debate since rTMS-evoked speech-errors appear relatively widespread over the brain and are rather poorly reliable, depending on the type of the speech-errors. We, therefore, investigated how rTMS-evoked speech-errors of distinct categories are located relatively to speech-related functional MRI (fMRI) clusters (serving as a widely used method for language mapping). Moreover, we compared three different rTMS-protocols with varying frequencies (10, 30 and 50 Hz). Methods 13 right-handed, healthy volunteers were investigated using fMRI (3T) and navigated rTMS using the same picture naming-task. To control for speech-related movement artefacts in fMRI, a sparse-sampling design was used with the following parameters: TR = 11 000 ms, delay in TR = 9000 ms, TE = 30 ms, flip angle = 90°, voxel size 3 × 3 × 3 mm3, FoV = 192 mm2, 79 images. After preprocessing (SPM8), the clusters of the superior temporal gyrus (STG) and the inferior frontal gyrus (IFG) were extracted. In the rTMS-sessions, 10 Hz, 30 Hz and 50 Hz rTMS were applied in a randomized order over the left hemisphere, continuously covering facial (pre-)motor and language-related cortical areas. Bursts were triggered to picture presentation without delay (picture-to-trigger interval = 0). Errors were rated by two independent raters using a post hoc video analysis and were categorized as follows: arrest, anomia, delayed term, complete delay, dysarthria, morphosyntactic errors, speech-motor disturbance, semantic and phonematic paraphasia (including neologisms). The coordinates of the sites (of maximal electrical field strength) corresponding to the speech-errors were extracted. Then the amount of rTMS-speech errors lying within the STG/IFG-clusters was calculated (relative to the total number of errors, “hit rate”) on an individual level using FSL. Results Overall, 17% of the rTMS-evoked speech-errors (“hit rate”) were located within the STG (10 Hz: 0.15 ± 0.05, 30 Hz: 0.19 ± 0.07, 50 Hz: 0.18 ± 0.07). In comparison, only 6% were located within the IFG (10 Hz: 0.06 ± 0.05, 30 Hz: 0.05 ± 0.02, 50 Hz: 0.07 ± 0.04). However, when corrected for the different cluster sizes (STG > IFG) this difference was not significant. Within the STG, a significantly higher rate of speech-errors could be found for 30 Hz as compared to10 Hz. Moreover, for motor-speech associated errors (e.g., delay) a tendency towards a higher rate of speech errors was found with higher rTMS-frequencies (30 & 50 Hz). Conclusion Our results suggest that rTMS language mapping is more sensitive for detecting language-relevant sites of the STG as compared to the IFG as identified via fMRI. This finding, however, results from the different cluster sizes. Within the STG, the number of speech-errors seems to increase with higher frequencies, especially for errors associated with motor-speech function. The results might differ for other speech related areas like the angular gyrus or primary motor areas, which needs to be further investigated.