We used motion-sensitive, recording accelerometers (Actiwatch™, Mini Mitter Company Inc., Sunriver, Oreg.) to estimate behaviors of Rocky Mountain elk (Cervus elaphus nelsonii). We placed rebroadcast LORAN-C radiocollars containing Actiwatches on 8 tame elk and recorded their observed behavior for 5,324 minutes over 27 trials in a natural setting. Concurrently, Actiwatches were summarizing and recording acceleration data at 1minute time intervals. We reduced behaviors to 3 categories: resting, feeding, and traveling. During trials elk spent 62.9% of the time resting, 36.3% feeding, and 0.8% traveling. Actiwatch readings were lowest during resting, higher during feeding, and highest when elk were traveling. Discriminant function analysis correctly classified 88% of the 1-minute intervals (98% for resting minutes, 72% for feeding, and 68% for traveling). Based on Actiwatch data, the estimated time in each trial devoted to resting, feeding, and traveling was correlated with actual, observed minutes in each behavior class (r = 0.95). We also tested the ability of Actiwatches to accurately estimate the most frequent behavior within 10-minute intervals. Using break-points determined by a discriminant function analysis of 1-minute data multiplied by a factor of 10, we successfully classified dominant behaviors 87% of the time.