The evaluation of immunohistochemistry (IHC) is usually semiquantitative, and thus subject to observer variability. We analyzed the reproducibility of different IHC measures. Fifty TMA cores of prostate cancer were stained for PDX-1, a transcription factor overexpressed in the cytoplasm of prostate cancer cells. The strongest intensity was scored 0-3 and 1-3 was used for extent (1-33%, 34-66%, and 67-100%). The stains were evaluated twice by four observers: two genitourinary pathologists, and two medical doctors with no formal pathology training. Staining intensity was also measured with automated image analysis. The pathologists read the slides faster than nonpathologists (total time 88 and 178 min, respectively, p = 0.03). Mean weighted kappa for intraobserver agreement was 0.85 (range 0.81-0.89) for intensity and 0.43 (range 0.38-0.51) for extent with similar results among pathologists and nonpathologists. Mean weighted kappa for interobserver agreement was 0.80 (range 0.77-0.84) for intensity and 0.21 (range 0.11-0.26) for extent. The subjective estimations of intensity correlated with results of image analysis (r = 0.61-0.66, p < 0.001), but the correlation between observers was stronger (r = 0.75-0.81) and correlated better with Gleason grade. Thus, subjective assessment of intensity can be done with a high level of reproducibility while estimation of staining extent is less reliable. Although educated pathologists were faster, the level of pathology training is not crucial for obtaining reproducible results in the analysis of TMA-based studies.