Effective use of the measured readings of sensors in on-line plant monitoring has been studied, based on an error theory. Both the measured reading and the calculated one, obtained by an analytical model of the plant, are treated as observed values, and the maximum-likelihood estimator is determined so as to minimize its mean-squared error. The difference between the estimator and the calculated reading is used to adapt the model to the current plant state and to increase accuracy of the calculated reading. The index of systematicness, which indicates the mutual independence of the two observed values, has been evaluated to determine the step in the procedure where the above adaptation is to be inserted. The error-theory-based model adaptation procedure has been experimentally applied to boiling water reactor power distribution calculations, and its performance has been verified in simulation calculations at different core states and different numbers of in-core neutron monitors by evaluating the expected error of the calculated readings. Compared to the adaptation, which uses the measured readings instead of the estimator, the error is typically lowered by more than 2% and is less affected by the number of monitors.