A solution to the state estimation problem under structural uncertainty (unknown or changeable dimension of the system state space) is given by the Interacting Multiple Model (IMM) filter. The requirements for its applicability under structural uncertainty are formulated. The highest IMM model probability is an indicator for the true model order and it can be used for structural identification. Results from test examples with stationary systems and systems with structural nonstationarity (changeable structure in the course of the time) demonstrate the filter efficiency. The scalar and multivariable cases are investigated.