The objective of this study on an ear-attached accelerometer in dairy cows was (1) to determine activity, rumination and lying time of the dams prior to calving, and include group level of measured variables (2) use the data to develop an algorithm to predict calving and (3) to test the performance of this algorithm. Video observations (24h/d) were used as reference for these events. Four weeks before expected calving, an ear-tag integrated tri-axial accelerometer (SMARTBOW system) was attached to the dam’s ear. A total of 894 data sets were analyzed from 2 weeks before estimated calving until the expulsion of the calf. The data set was divided into training, validation and test data sets. Input variables based on accelerometer data were used to develop an algorithm for calving prediction and included rumination time, activity level and lying time of individual animals and on group level. To model the upcoming calving on the basis of the input variables, Transfer Function models were used. The performance of the algorithm to predict calving in the time periods of 72, 48, 24, 12, 6, 3 and 1 h before expulsion of the calf was determined. Greatest accuracy (94%) was achieved at -12 h and maintained on this level until the expulsion of the calf. Greatest sensitivity was achieved at -1 h (54%), lowest sensitivity was found at -72 h (19%). Our findings indicate that accelerometer-based prediction of parturition with the SMARTBOW system using the described input variables is eligible to support calving management in dairy cows. Further studies should test and improve the algorithm under different farming and management systems.