The onset of spring melt and autumn freeze-up of Arctic sea ice are shown, through comparison with coincident ERS-1 synthetic aperture radar (SAR) and surface air temperature data, to produce significant and detectable changes in the passive microwave signature. Melt onset is marked by a sharp rise in brightness temperatures resulting from the increased emissivity of moisture in the snow cover. Freeze-up is accompanied by a sharp drop in brightness temperatures because of volume scattering by bubbles in the upper ice layer. An algorithm is developed to detect melt and freeze-up from passive microwave data and is applied to the Scanning Multichannel Microwave Radiometer (SMMR) and Special Sensor Microwave/Imager (SSM/I) data from 1979 to 1996. The mean accuracy of melt and freeze-up dates determined from the SSM/I is estimated to be ¿2 and ¿3 days, respectively. The reduced temporal sampling of the SMMR degrades the accuracy by 1 day. On average, melt onset varies from late May in the southern Beaufort Sea to mid-June in the central Arctic north of the Canadian archipelago, while freeze-up varies between mid-September and mid-August. The shortest melt season appears to occur north of the Canadian archipelago rather than at the North Pole, which is consistent with the fact that minimum summer air temperatures are centered over Greenland. The mean standard deviations of the interannual variability of melt and freeze-up are 6.8 and 11.4 days, respectively. ¿ 1998 American Geophysical Union