How are complex words recognized during the early moments of visual word recognition? What roles do full word and constituent frequency play in semantic processing? The present study addressed these questions by employing a word-picture relatedness task with brief stimuli presentations designed to tap the early mapping of orthographic input onto semantic representations. The main manipulation involved first presenting a picture depicting the target word’s constituent (200 ms), followed by the presentation of the target word (56 ms). We compared the rate of positive relatedness judgements elicited by picture-word pairs between suffixed (SKI-skier), pseudo-suffixed (MOTH-mother), and non-suffixed words (CAN-canoe). Results suggest that the “constituents” of all three word types are semantically accessed, although with a suffixed word advantage. Regression analyses did not corroborate behavioral findings as no full word and constituent frequency effects were obtained. We discuss the implications of these findings for models of the visual word recognition system.