Most Chinese compound characters represent meaning explicitly and systematically via the semantic radical. Previous research has shown that readers can use the semantic radical to infer the meaning of unknown compound characters (i.e., semantic decoding). Yet it is unclear whether it influences reading in natural texts or facilitates orthographic learning. Across two eye-tracking experiments, we exposed adult native speakers (n= 25 and n=16) to sixteen pseudocharacters embedded in sentences. Half of the pseudocharacters contained semantic radicals related to the meaning conferred by the sentential context (i.e., transparent); the other half were unrelated (i.e., opaque). After reading, participants completed written cloze, orthographic decision and definition production tasks. In both experiments, eye-tracking results showed a reduction in fixation times and regressions over the five exposures for the pseudocharacters. The post-test outcomes provided evidence for orthographic and semantic learning. There was no difference in eye-movement measures or in written cloze and orthographic decision between transparent and opaque items, but definition was better for the transparent pseudocharacters. We conclude that semantic decoding does not affect the reading of novel compound characters in natural texts or orthographic learning, but that it does seem to assist in learning semantics.