Dynamic body acceleration (DBA) derived from the data obtained by tri-axial accelerometer has been widely used in various animal species as a proxy for estimating energy expenditure (EE) under free-moving conditions. More recently, the use of 9-axis sensor equipped with tri-axial accelerometer, tri-axial gyroscope and tri-axial magnetometer is examined to estimate EE as well as tri-axial accelerometer. In order to estimate EE correctly from DBA, sensors should be placed near the center of mass of animals. However, in practice, livestock farmers attach sensors usually on collar because the place is easier to be attached. Therefore, the aims of this study were to examine the relationship between the sensors attached on back and collar of animals and to develop EE estimation models from the sensor data attached on collar in cattle and sheep. Four pregnant Japanese Black cows and four castrated Corriedale sheep were used in this study. Twenty-four hours of sensor data in grazing conditions were collected. First, the correlations of the sensors data attached on back and collar were analyzed by each behavioral classification. Then, the models to estimate the calculated EE derived from DBA gained from the sensor placed on back were developed by regression analyses by cows and sheep. The independent variables in the models were behavioral indicators calculated from 9-axis sensors attached on collar and the dependent variables in the models were calculated EE from the sensor data on back. As a result, the distinctive relationship of behavioral indicators between collar and back in each livestock species were found. As for the development of EE estimation models, although the effect of animals in the model was significant in each species, it was found that the models without the effect of animals could also estimate EE with some degree of accuracy; there is a possibility to estimate EE with the indicators derived from the sensor attached on collar. Also, the coefficients of determination (R2) of the models using 9-axis sensor data from collar were similar with those of the model using only tri-axial accelerometer data from collar, indicating that the use of tri-axial accelerometer data is enough for EE estimation of grazing livestock animals.