Recently, various kinds of healthcare systems for the elderly have been developed. Falling detection is one of the important tasks to protect them from crucial accidents. Cameras, acoustic sensors, and accelerometers are mainly used to detect the falling. However, from the viewpoint of false alarm rate, privacy issues, and intrusiveness of the devices, each method has its own shortcomings. Doppler sensor is a palm-sized device, and can be implemented for highly accurate human activity recognition without wearable sensors. Doppler sensor is less sensitive to the movements orthogonal to the irradiation direction. Thus, a method to compensate this characteristic is needed. We propose falling detection using multiple Doppler sensors to raise the precision of falling detection covering the multi-directions of the target movement. Two or three sensors are exploited, and the extracted sensor data is processed by a feature combination or selection method. The resulting data are classified by support vector machine (SVM) or k-nearest neighbors (k-NN). We evaluate several kinds of falling, "Standing - Falling," "Walking - Falling," and "Standing up - Falling," and non-falling like "Walking," "Lying on floor," "Picking up," and "Sitting on a chair." These activities are tested toward 8 directions spaced at respective intervals of 45 degrees. The results show that the combination method, using three sensors, achieves 95.5 % accuracy of falling detection, and the selection method, using three sensors, achieves 93.3 % accuracy. We also discuss the accuracy of each activity direction and the viability of these methods for the practical use.