Quality control of sputum smear microscopy, which is essential for ensuring correct tuberculosis (TB) diagnosis, is often performed through the unblinded rereading of all positive slides and a sample of negative slides. To assess misclassification error introduced by knowledge of prior results. The Southern Vietnam Regional TB Laboratory prepared three gold-standard sets of 750 slides: an unblinded set, an unblinded set in which 13% of negative slides were replaced by weakly positive slides purposefully mislabelled as negative, and a blinded set. Six provincial technicians who normally perform district quality control each reread 125 slides from each set. In the three sets only one negative slide was misread as positive. In the unblinded set (referent), 2.9% (9/311) positive slides were misread as negative, compared with 18.7% (57/305) in the blinded set (prevalence ratio [PR] = 6.5; 95% confidence interval [CI] 3.3-12.8; P < 0.001), and 11.3% (33/293) in the unblinded set with mislabelled slides (PR = 3.9; 95%CI 1.9-8.0; P < 0.001). False-negative error was more common than false-positive error. Knowledge of prior reading influences re-reading. Blinded re-reading of systematically selected slides would appear preferable, although this method requires high levels of proficiency among quality control technicians.