Four membrane filter methods for the enumeration of fecal coliforms were compared for accuracy, specificity, and recovery. Water samples were taken several times from 13 marine, 1 estuarine, and 4 freshwater sites around Puerto Rico, from pristine waters and waters receiving treated and untreated sewage and effluent from a tuna cannery and a rum distillery. Differences of 1 to 3 orders of magnitude in the levels of fecal coliforms were observed in some samples by different recovery techniques. Marine water samples gave poorer results, in terms of specificity, selectivity, and comparability, than freshwater samples for all four fecal coliform methods used. The method using Difco m-FC agar with a resuscitation step gave the best overall results; however, even this method gave higher false-positive error, higher undetected-target error, lower selectivity, and higher recovery of nontarget organisms than the method using MacConkey membrane broth, the worst method for temperate waters. All methods tested were unacceptable for the enumeration of fecal coliforms in tropical fresh and marine waters. Thus, considering the high densities of fecal coliforms observed at most sites in Puerto Rico by all these methods, it would seem that these density estimates are, in many cases, grossly overestimating the degree of recent fecal contamination. Since Escherichia coli appears to be a normal inhabitant of tropical waters, fecal contamination may be indicated when none is present. Using fecal coliforms as an indicator is grossly inadequate for the detection of recent human fecal contamination and associated pathogens in both marine and fresh tropical waters.