Abstract: The determination of observation weights and error models in levelling networks is a current problem. The method of Minimum Norm Quadratic Estimation (MINQE) is proposed for the determination of the error model. Different possible models of error structure of levelling measurements are estimated from the adjustment of a network or from the forward-backard discrepancies of levelling lines using the iterated MINQE. An appropriate model of error structure can be obtained after comparing the estimated parameters in the models with their standard deviations. As an example, a first-order levelling network has been analysed. The results demonstrate the usefulness of the method.