Two experiments compared how relatively long novel prefixed words (e.g., "overfarm") and existing prefixed words were processed in reading. The use of novel prefixed words allows one to examine the roles of whole-word access and decompositional processing in the processing of non-novel prefixed words. The two experiments found that, although there was a large cost to "novelty" (e.g., gaze durations were about 100 ms longer for novel prefixed words), the effect of the frequency of the root morpheme on fixation measures was about the same for novel and non-novel prefixed words for most measures. This finding rules out a ("horse-race") dual-route model of processing for existing prefixed words in which the whole-word and decompositional route are parallel and independent, as such a model would predict a substantially larger root frequency effect for novel words (where whole-word processes do not exist). The most likely model to explain the processing of prefixed words is a parallel interactive one. (Contains 4 tables and 4 footnotes.)