Building deformation mode identification, which is important for evaluating safety in structural health monitoring applications following seismic events, may be difficult to perform if there are some floors without sensors. This study proposed potential features based on the response of monitored floors which may indicate the correct deformation mode. These features were evaluated against shake-table test data of steel frames and numerical analyses of RC frame buildings. The features were found to be reasonable and can be used in machine-learning applications to identify deformation modes.