The immunogold method is widely used to localize, identify, and distinguish cellular antigens. There are, however, some pitfalls that can lead to nonspecific binding, particularly in cytoskeletal studies with gold probes prepared from small gold particles. We present a list of suggestions for minimizing nonspecific binding, with particular attention to two problems identified in this study. First, we find that the method used to prepare the colloidal gold particles affects the degree of nonspecific binding. Second, the standard BSA-stabilized small gold probes evidently possess exposed regions that bind to the proteins of cytoskeletal preparations. This was investigated in whole-mount cytoskeletal preparations of cultured cells by use of light microscopy, transmission electron microscopy, and photoelectron microscopy of silver-enhanced specimens. Gold probes were made from approximately 5-nm particles generated by reduction of HAuCl4 with three different reducing agents: white phosphorus, sodium borohydride, and citrate-tannic acid. All three preparations stabilized in the conventional way showed significant levels of nonspecific binding, which was highest with citrate-tannic acid. This problem was largely solved with all three types of probes by including fish gelatin in the probe buffer, by substituting fish gelatin for the BSA stabilizer used to prepare the probes, or by pre-adsorption methods. Application of these techniques resulted in clear immunogold labeling patterns with minimal nonspecific background.