This paper reports on a series of three studies designed to assess the effectiveness of a metacognitive approach to teaching word identification and reading comprehension skills to upper primary poor readers, and to investigate effective methods for implementing the metacognitive program in the regular classroom. To improve word identification skills, experimental subjects were trained to: consider the context; compare with known words; and carve up the word parts. To help monitor and control the use of these strategies, subjects were taught to: look for the cues; be flexible; and ask does it make sense. Reciprocal teaching procedures, incorporating the word identification strategies, were used for comprehension training. Subjects in control conditions received: (1) reciprocal teaching of comprehension skills and traditional methods of word identification; (2) normal classroom reading activities; or (3) normal classroom activities plus phonics-based remedial instruction. Measures of word identification, metacognitive awareness and monitoring of word identification cues, and comprehension were taken on several occasions in each study. Results of repeated measures analysis of variance showed significant improvements in most measures for all conditions. However, there were significantly greater improvements for subjects in the experimental conditions. Also, a model of implementation in which teachers were entirely responsible for implementation was more effective than one in which the experimenter initially set up the program and the teachers gradually took over responsibility for its implementation. The implications of these findings for classroom practice are discussed in the light of current research. (Contains 40 references and 11 figures of data.) (Author/RS)