A new tool is proposed for finding out the completeness limit in apparent magnitude of a magnitude-redshift sample. The technique, closely related to the statistical test proposed by Efron & Petrosian, presents a real improvement compared to standard completeness tests. Namely, no a priori assumptions are required concerning the redshift-space distribution of the sources. It means in particular that neither the clustering nor the evolution of the mean number density of the galaxies affects the result of the search.