We compare poor-performing and normal-performing decoders' processing times on real words, pseudo-homophones, and nonwords (Study 1), and evaluate how a processing time difference is associated with rates of decoding development (Study 2). Over 800 sixth and seventh graders took an online reading component battery, which included a decoding test, four times in three consecutive years. Study 1 indicates that poor decoders were generally "slower" than their peers in processing real words and pseudo-homophones, but they spent "shorter" time than their peers when decoding nonwords that were novel to them, resulting in a significant interaction. In Study 2, longitudinal modeling reveals that the time students spent decoding novel words positively predicted decoding development. These results are consistent with the hypothesis that poor decoders may be trapped in a vicious cycle: poor decoding skill combined with less time spent attempting to decode novel words interferes with decoding development.