The latest version of the connectionist dual process model of reading (CDP++.parser) was tested on a set of nonwords, many of which were orthographically strange (e.g., PSIZ). A grapheme-by-grapheme read-out strategy was used because the normal strategy produced many poor responses. The new strategy allowed the model to produce results similar to individual participants. Given that the strategy uses only simple information acquired in the normal course of learning, it shows that it is a plausible method for reading when the normal method fails and offers insight into how groups with low orthographic exposure might read.