Ample evidence has shown facilitations of context-based prediction on language comprehension. However, the influential effect of working memory capacity on this predictive processing remains debated. To investigate this issue with the electroencephalograph technique, high and low working memory capacity participants read strong-, moderate- and weak-constraint sentences which resulted in high-, moderate- and low-predictability for the critical nouns. The strong-constraint (vs. weak-constraint) contexts preceding the nouns elicited a larger positive deflection, which was only observed for the high-span group. Along with the smaller N400s for strong- vs. weak-predictable nouns for both groups, the moderately predictable nouns elicited smaller N400 than the weakly predictable nouns for the high-span group. The ERP effects at both verbs and nouns correlated significantly with the noun's predictability. These findings suggest that predictive processing involves at least partially an effortful-meaning-computation mechanism, and high working memory capacity facilitates the activation and integration of predicted information during language comprehension.