To improve accuracy of navigation the ZUPT (zero velocity update) technic is commonly used, aiding the IMU with information on zero foot velocity during the stance phase of a step. This information is fed to a extended Kalman-type filter. The state vector of the filter usually contains position, velocity and orientation of the IMU. We show that the ZUPT condition can be written in two ways of which the most commonly used yields inconsistent results. For that purpose we employ covariance analysis. We suggest a decomposition of the error equations into the so-called dynamic and kinematic errors and decompose these equations into four simple subsystems. For each subsystem, error covariance can be written in explicit formulas.