The comparison between spoken and handwritten production in picture naming tasks represents an important source of information for building models of cognitive processes involved in writing. Studies using this methodology systematically reported longer latencies for handwritten than for spoken production. To uncover the origin of this difference across modalities, we compared the latencies of spoken picture naming and two written picture naming conditions: one in which the participants could see and monitor their handwriting ("visible-condition"), and one in which they could not monitor their production ("masked-condition"). Previously reported differences between spoken and handwritten naming latencies were replicated in the standard visible-condition. By contrast, production latencies were faster in the written masked-condition than in the visible-condition and did not differ from spoken production latencies. These results suggest that longer handwriting latencies, in comparison with speaking latencies, are due to the delayed onset in handwriting in conditions where the sheet is visible. The implications of these results on both written production models and experimental methods are discussed.