Understanding the spatio-temporal elimination pattern of grazing cattle is important for grazing management. We thus developed a new method of detecting defecation events using a three-axis accelerometer. The accelerometer was fixed on the tails of three Japanese Black steers in a pasture, with the x-, y-, and z-axes being set to the front-to-back, side-to-side, and vertical directions relative to the normal tail position, respectively. The defecation behavior was also visually observed. The 3-sec moving average was calculated from raw acceleration data and charted along the time course. The x-axis and z-axis accelerations showed convex upward and downward curves, respectively, at the defecation events. By using the synchronous signs of both curves, we could visually detect virtually all defecation events. And in order to detect defecation events automatically, we created six variables (i.e., maximum, minimum, and area in convex curve per 30 sec for x- and z- axes) and applied quadratic discriminant analysis (QDA) and a support vector machine (SVM). The critical success index values in QDA and the SVM were 0.8 and 0.98, respectively, using the leave-one-out cross-validation method. We concluded that the use of an accelerometer on a steer’s tail is effective in visually and statistically detecting defecation events.