To understand the causes of differences in language ability we must measure the specific and separable processes that contribute to natural language comprehension. Specifically, we need measures of the three language subsystems – semantics, syntax, and phonology – as they are used during the comprehension of real speech. Event-Related Potentials (ERPs) are a promising approach to reaching this level of specificity. Previous research has identified distinct ERP effects for each of the subsystems – the N400 to semantic anomalies, the Anterior Negativity and P600 to syntactic anomalies, and the Phonological Mapping Negativity to unexpected speech sounds. However, these studies typically use stimuli and tasks that encourage processing that differs from real-world language comprehension. Further, previous ERP studies indexing language processing in young children not only use unfamiliar tasks, but also typically exclude data from the large proportion of children. We need to measure language-related ERPs in a context as close as possible to real-world processing, and in a manner that includes data from representative rather than highly-selected samples of children. The experiments described in this dissertation achieve that goal. Adults and five-year-old children listened to a child-directed story while answering comprehension questions. Infrequent violations were included to independently probe the three language subsystems. In children and adults, the canonical N400 response was evident in response to semantic violations. Morphosyntactic violations elicited a long-duration Anterior Negativity without a later P600. Phonological violations on suffixes elicited a Phonological Mapping Negativity in adults. This is the first report of this phonological effect outside of highly-predictable lexical contexts. Popular normed behavioral assessments were also administered to the children who participated in this study. Results from these assessments confirmed that performance on tasks claiming to measure categorically different abilities are correlated with one another, and that language measures correlate with so-called nonverbal measures. ERPs indexing different language subsystem did not correlate with each other or with measures of nonverbal cognitive ability. Using multiple ERP measures during natural language comprehension, we are able to isolate specific aspects of language processing, increasing the possibility of making meaningful connections between biology, experience, and resulting language ability.