An experiment examined whether metalinguistic awareness involving the detection of semantic ambiguity can be taught and whether this instruction improves students' reading comprehension. Lower socioeconomic status third graders (M age = 8 years, 7 months) from a variety of cultural backgrounds (N = 46) were randomly assigned to treatment and control groups. Those receiving metalinguistic ambiguity instruction learned to analyze multiple meanings of words and sentences in isolation, in riddles, and in text taken from the Amelia Bedelia series (Parish, 1979, 988). The control group received a book-reading and discussion treatment to provide special attention and to rule out Hawthorne effects. Results showed that metalinguistic ambiguity instruction was effective in teaching students to identify multiple meanings of homonyms and ambiguous sentences and to detect inconsistencies in text. Moreover, this training enhanced students' reading com prehension on a paragraph-completion task but not on a multiple-choice passage-recall task, possibly because the two tests differ in the array of linguistic or cognitive correlates influencing performance. Comprehension monitoring was not found to mediate the relationship between ambiguity instruction and reading comprehension. Results carry implications for the use of language-based methods to improve reading comprehension in the classroom.