Objective. The aim of this study was to determine the strength of the correlation between colposcopic impression and biopsy histology. Methods. In an urban referral clinic, colposcopy and directed biopsy were performed between July 1, 1996, and December 31, 1999, by residents supervised by board-certified attending obstetrician-gynecologists. Impression and biopsy were graded as benign, suggesting condyloma or koilocytosis, cervical intraepithelial neoplasia (CIN) grades 1-3, or cancer. The significance of association was assessed by χ2 testing and the strength by κ statistics. Results. Colposcopies were performed on 2825 women, with colposcopic impression and biopsy grade known for 2112. Exact agreement was found in only 893 (37%) women, but results agreed within one grade in 1203 (75%). The association between impression and histology was significant (P < 0.001), but the strength of the correlation was poor (0.20). The positive predictive value of any colposcopic abnormality for any histologic abnormality was 80%. The negative predictive value of a benign colposcopic impression was 68%. The sensitivity of colposcopy with a threshold of any lesion detected was 89%, and the specificity was 52%. The sensitivity for CIN 2/3 was 56%. Conclusion. Colposcopy is imprecise, although useful in estimating lesion grade. Management decisions require biopsy.