Certification tests currently employed by some industrialized nations to certify respirators require that when challenged with air flow at a constant suction head of 25 mmH2O, the leakage into the facepiece from the exhalation valve(s) should not exceed 30 ml min(-1). However, the test alone might not reflect the leakage rate under different levels of vacuum. To study the characteristics of leakage through exhalation valves, a leakage meter was built to measure the leakage rate of four brands of exhalation valves. Seven valves of each brand were tested for leakage rate at pressure drops ranging from 15 to 45 mmH2O (or extended to 75 mmH2O for some valves). Two types of leaks, i.e. fiber insertion and arch dent on the valve seat, were used to investigate the effect of leak shape on the characteristic leak flow as a function of pressure drop. A leakage data point represented a mean value of five measurements. The results showed that the leakage rate could increase or decrease with increasing pressure drop inside the respirators, indicating that the current method of conducting leakage tests, at a constant suction head of 25 mmH2O, does not guarantee better protection than when the respirators are used at a lower suction head. Moreover, based on the sample valves tested in this study, it was found that valve leakage was mostly caused by defects in the valve seats. Both fiber-insertion and arch-dent leaks had either increasing or decreasing characteristic curves, depending on the size of the leak and the material of the valve membrane.