The purpose of this study was to evaluate the reliability of different methods for estimating neovascularization in breast cancer and to compare them in terms of observer variability. The microvessel endothelium was stained immunohistochemically by antibodies against CD34. The investigated methods included Chalkley counting, estimation of intratumoral microvessel density (MVD) by one hot-spot, MVD by the mean value of three hot-spots, and the highest value of MVD in three hot-spots. In addition, we applied stereology in the quantification of angiogenesis in the whole tumor section by random and systematically distributed sampling fields. Each of forty tumors was measured with all methods, twice by the same observer and once by another observer. Observer variation was analyzed by orthogonal regression, estimating the slope and intercepts with 95% confidence intervals (CI), and by analysis of agreement using difference plots. Intraobservationally, the methods had variations of the same magnitude (coefficient of variation [CV] approximately 20%). Interobservationally, the stereologic estimate of vessel profiles, Q(A), from the whole tumor section and the Chalkley counting method had the lowest variation (CV approximately 21%), with a small contribution by observers alone (CV 8% to 9%). Interobservationally, the MVD methods had considerable variation with a large contribution by observers alone (CV approximately 30%), which was lowest using the mean of three hot-spots. Correlation slope and 95% CI of Chalkley were 1.18 (0.95, 1.48), CV 20%; slope of MVD (mean) was 1.14 (0.91, 1.43), CV 31%; and slope of MVD (max) was 1.15 (0.92, 1.45), CV 36%. The slope of MVD on one hot-spot was 1.33 (1.08, 1.63); CV 38%. Additional measurements performed using a conference microscope, eliminating subjectivity in hot-spot selection and field sampling, optimized the reproducibility: slope was 1.02 (0.99, 1.04); CV of differences, 3.5%. On the other hand, reproducibility was not necessarily optimized by choosing the same hot-spot area, because variation in selecting a microscopic field could yield different counting numbers. The stereologic estimation of QA based on the whole tumor section had a high reproducibility, with low variation due to observers. The Chalkley and MVD methods had moderate reproducibility, and the Chalkley method had low variation due to observers alone. For all methods, the biologic variation among patients was the major contributor to the total variation. The Chalkley and MVD methods have been published to provide significant prognostic estimates in breast cancer, but the Chalkley method has less observer variation and may be superior from a methodologic point of view.