Auroral roar, a radio emission occurring near the second and third electron cyclotron harmonics, has been monitored during a 51-day interval in 1998 from three ground-based high-latitude observatories: Churchill (69¿ corrected geomagnetic latitude (cgm)), Baker Lake (74¿ cgm), and the Early Polar Cap Observatory (EPCO) at Resolute (84¿ cgm). These observations reveal a latitude dependence of the local time distribution of auroral roar events. In the auroral zone they occur predominantly in the premidnight-midnight sector as previously reported, but in the polar cap they occur evenly throughout nighttime hours. Sorting the data by magnetic activity level confirms the effect; for example, the distribution of events at Baker Lake during times of extremely low activity (Kp<1) is concentrated near magnetic midnight, consistent with Baker Lake lying in the auroral zone when the oval is contracted. The latitude dependence of the local time distribution may be explained by propagation effects, with inefficient propagation in the auroral zone causing only relatively locally generated emissions to be detected there, while stations well poleward of the auroral zone detect emissions from sources widely distributed in longitude and latitude. Statistics of the frequencies of auroral roar events detected at Resolute show evidence of an upper frequency cutoff near 2960 kHz. Ray-tracing calculations suggest that it is plausible for auroral roar to reach Resolute from sources up to 1500 km distant in the auroral zone, but they do not provide an obvious explanation for the frequency cutoff. ¿ 2001 American Geophysical Union