Previous studies show that 7–8-year-old poor comprehenders differ from good comprehenders, matched in age and decoding skill, primarily in their failure to make high-level inferences, despite adequate text recall. The impact of inference awareness training on reading comprehension in two such groups was compared with the effects of two other treatments. Inference-trained children were instructed over 4 weeks in making inferences from text and generating questions. Other groups were trained either in rapid decoding or in standard comprehension exercises. Less skilled comprehenders given inference training improved significantly more than those given decoding practice, and slightly, but not significantly, more than those given comprehension exercises. Skilled comprehenders showed little improvement regardless of treatment condition. The results are discussed in relation to possible sources of comprehension failure, and implications for remediation.