Compositional data sets usually contain zero values, which may or may not correspond to the genuine absence of a component. The most common form of zeros are those resulting from insufficiently large samples and insufficiently accurate instruments. These zeros are an impediment in the analyses because the log-ratio transformations require positive data matrices. In recent years, several techniques have been proposed for addressing the ``zero problem'', all highly dependent on the nature of the zeros. A required condition for any approach is coherence with the vectorial structure of the simplex. In this paper these methodological aspects are revisited and illustrated using compositional data from nutritional science, geochemistry, ichthyology, and psephology.