Micro-milling is prominent among other micro-manufacturing processes due to their abilities in manufacturing of 3D features, high material removal rates and high precision. One of the most important challenges of this process is tool deflections which contribute even up to 90% of dimensional errors of the finished product. This paper addresses a novel method to estimate micro-milling tool deflections applicable in micro-milling machines equipped with linear motors. In this method, position feedbacks and inputs to the amplifiers are used to real-time estimation of cutting forces by applying Kalman filter. Outputs of the estimator include a resultant of all disturbing forces in servo control loop of the motors. Therefore, cutting forces need to be compensated for other disturbing forces that are mostly friction and force ripples in linear motors. To compensate them, neural networks were used. A neural network with a hidden layer and 16 nodes inside, and with two time-delayed lined (TDL) could well model friction and force ripples. Results showed that the proposed tool deflection method is able to estimate 22% of micro-milling tool deflections.