The invention relates to a method for computing a bound B up to a given confidence level 1−α, of an error in a state vector estimation KSV of a state vector TSV of a physical system as provided by a Kalman filter. The method decomposes the errors of the Kalman solution as a sum of the errors due to each of the measurement types used in the filter, In addition, the contribution of each type of measurement is bounded by a multivariate t-distribution that considers the error terms from all the epochs processed. Then, the method implements three main operations: computing a probability distribution of the measurement errors for each epoch and measurement type; summing the previous distributions to obtain a global distribution that models the Kalman solution error; and computing the error bound B for a given confidence level from the resulting distribution.