Reliable antibodies represent crucial tools in the arsenal of the cell biologist and using them to localize antigens for immunocytochemistry is one of their most important applications. However, antibody-antigen interactions are much more complex and unpredictable than suggested by the old 'lock and key' analogy, and the goal of trying to prove that an antibody is specific is far more difficult than is generally appreciated. Here, we discuss the problems associated with the very complicated issue of trying to establish that an antibody (and the results obtained with it) is specific for the immunolabeling approaches used in light or electron microscopy. We discuss the increasing awareness that significant numbers of commercial antibodies are often not up to the quality required. We provide guidelines for choosing and testing antibodies in immuno-EM. Finally, we describe how quantitative EM methods can be used to identify reproducible patterns of antibody labeling and also extract specific labeling distributions.