Language comprehension involves the integration of semantic, syntactic, and pragmatic sources of knowledge. The mechanisms and time series of the integration process are a major issue in psycholinguistic research. In this study we investigated whether Japanese native speaker use parallel or serial models when they read Japanese sentences. While parallel models pose that semantic and syntactic information are processed independently, serial models posit that semantic and syntactic information are processed serially and dependently each other. We recorded event-related potentials (ERPs) with four types of sentence including correct (Correct), semantic mismatch (SemMM), syntactic mismatch (SynMM), and semantic and syntactic mismatch (Sem_SynMM) conditions. Results showed that larger N400 (360-460 ms) arose in a dual mismatch condition than in semantic or syntactic conditions. In addition, we observed later negativity (580-640 ms) only in the Sem_SynMM condition. Our results suggested that, though most of the previous studies supported syntactic first serial model, temporal relation between semantic and syntactic processing could depend on different languages, tasks, and grammatical violations using in the experiment.