In this paper an Extended Kalman Filter (EKF) is used in order to estimate the real state of a humanoid robot (HRP-2 robot in our case study) using the combination of the information coming from the encoders (kinematics) and from the Inertial Measurement Unit (IMU). The integration of the kinematic information into the Kalman filtering process allows a good estimation of the attitude and reduces the complexity of the problem to the use of simple kinematic transformations, even considering the existence of accelerations and mechanical flexibilities in the robot. The EKF estimator presented here is an open solution directly applicable to any humanoid robot, which is the main contribution of our approach. Experimental results are given showing the good performance of the method.