Crackles are commonly used in clinical decision-making, and in certain diseases the number of crackles reflects the severity of the illness. Auditory crackle estimations are subjective; crackle counting from time amplitude plots of sound (called time-expanded waveforms) is more objective but is cumbersome. We devised a computer-based system to count crackles automatically. One hundred samples of lung sounds from 41 subjects were recorded using an electret microphone air-coupled to the chest wall. Interobserver agreement in estimating the number of crackles per breath was high (r = 0.88, p less than 0.001), and these counts were significantly correlated with counts made of spikelike deflections seen on time-expanded waveform analysis (r = 0.78, p less than 0.001). The automatic crackle counting correlated with the physician counts (r = 0.74, p less than 0.001). The average number of crackles counted per breath was greater by visual inspiration (8.8) and by automatic analysis (7.8) than it was by the physician observers (5.8). Reasons for the discrepancies include the fact that there are no absolute criteria for crackles and that rapidly occurring crackles are difficult to count by ear. Counting crackles by computer-based methods is feasible and can improve noninvasive cardiopulmonary diagnosis.