Although many researchers have reported high efficacy rates of some computer-aided diagnosis (CAD) schemes, it is known that their performance depends strongly on the database used for the tests. This is an important task to the comparison of different schemes performance, since they usually are developed and tested with a particular images set. Previously, we have reported the development of an image processing scheme designed to detect clustered microcalcifications as a part of a CAD scheme. Therefore, in this work we are reporting the performance results of such a processing procedure after tests with three different images databases: two corresponding to mammograms obtained from Hospital das Clinicas de Ribeirão Preto, Brazil, but from different units and digitized in different scanners; and a third corresponding to mammographic images obtained directly in digital form by Internet from the National Expert and Training Centre for Breast Cancer Screening at the University of Nijmegen, the Netherlands. As expected, the scheme efficacy in detecting clusters has changed according to the images set tested: from 95% of efficacy (true positive plus true negative results) for the best situation to 88% for the worst one. In addition, we discuss briefly some factors relative to the images sets which can change a CAD scheme result.