With the assistance of his therapist, each participant constructed a written vignette that described his most recent or typical offense. The vignette was recorded by the offender in the present tense on an audiotape. Each vignette was divided into four sections that corresponded to the outline proposed by Pithers et al. (1983). Segment 1 described background issues; segment 2 comprised a high-risk situation where the participant's sense of self-control over sexual behavior was threatened; segment 3 contained the lapse, which is defined as a behavior that occurred before the actual offense that reflected an intent to offend; segment 4 described the first instance of any sexual behavior. In playing back the audiotape for each offender, he was asked to give a brief summary of the thoughts and feelings experienced during the segment just heard. A second study provided cross-validation and reliability of classification estimates of the categories developed in the first study. The resulting descriptive model consists of nine stages, with three contributing factors that describe the sequence of cognitive and behavioral events that form an offense chain. The model incorporates offender type, offense type, and a description of the possible interactions between the various stages and factors. The model was then applied to the offense descriptions of an independent sample of 12 incarcerated child molesters, so as to determine its cross-sample validity and the reliability of classification. The results suggest that the model has provisional validity and adequate interrater reliability. The theoretical, research, and clinical implications of the descriptive model are discussed. 1 figure and 23 references