Two methods of surveillance for nosocomial infection--routine notification and active case finding--were compared during three consecutive months at the high risk newborn unit of the Hospital de Base of the Federal District, involving 66 newborns, which represented 93% of the total patient population referred to that unit for specialized care. Our purpose was to measure the efficiency of routine notification. While this method involved infection notification forms routinely filled by the physician at discharge time, active surveillance consisted of physical examination and chart review of all the newborns, carried out, independently, twice a week: routine notification estimated a 27.3 per cent prevalence rate, versus a 30.3 per cent by active-case finding, considered by the authors a superior surveillance method. Although infection rates were similar, routine surveillance resulted in many false-positive and false-negative notifications, reaching a sensitivity of 60.0 per cent and a specificity of 86.9 per cent. The resulting positive predictive value was 66.7 per cent and negative, 83.3 per cent. The data collected by routine notification was often incomplete and inconsistent. In conclusion, the results seem to indicate that control programs based solely on routine surveillance could be producing inaccurate nosocomial infection rates and, consequently inadequate control measures.