The 20 x 10(9)/L (20,000/microliters) threshold for prophylactic platelet transfusion may be unnecessarily high. The widespread use of this threshold may reflect lack of confidence in the reliability of low platelet counts. We evaluated the performance of automated platelet counts and their relation to clinical bleeding. First, we prepared serial blood dilutions with "target" platelet counts from 2 to 40 x 10(9)/L. For the 48 measurements on 2 x 10(9)/L "target" dilutions, values of 1 or 2 x 10(9)/L were obtained with the Sysmex NE-8000 analyzer (mean 1.44 x 10(9)/L; SD 0.31 x 10(9)/L). Similarly, for 5 x 10(9)/L "target" counts, automated counts were 3-6 x 10(9)/L (mean 4.42 x 10(9)/L; SD 0.18 x 10(9)/L). Similar results were observed with all other "target" levels, with coefficients of variation (CV) from 6.39% to 7.71% with 10-40 x 10(9)/L "target" values. Secondly, we compared triplicate automated and manual platelet counts on thrombocytopenic patients with platelet counts from 4-30 x 10(9)/L. The triplicate automated platelet counts differed by no more than 5 x 10(9)/L among themselves, whereas the manual counts varied by as much as 30 x 10(9)/L. Mean platelet counts: automated, 14.40 x 10(9)/L (CV 10.12%); manual, 16.48 x 10(9)/L (CV 30.39%) (P = 0.038 for counts; P < 0.001 for CV). Finally, we prospectively evaluated bleeding in thrombocytopenic patients (1,809 patient-days of observation). Univariate and multivariate logistic regression analysis revealed highly significant correlations between the automated platelet count and major and minor bleeding manifestations. Thus, automated platelet counts are highly reliable and accurately predict clinical bleeding. The use of automated analyzers should facilitate improved prophylactic platelet transfusion protocols.