This paper presents a bi-polynomial least squares approximation method (LSM) related to the Callendar-Van Dusen (CVD) model to analyse the estimation accuracy. A Kalman filter (KF) is used as a possible alternative for estimations in the presence of noise in the measurements. It is shown that the KF obtained a higher estimation accuracy of the sum of squared residuals (SSR) index with respect to the least squares method (LSM), which minimises the squared estimate of errors (SSE). Thanks to these results, it is possible to choose between LSM or the KF for an adequate fitting using SSE or SSR. A case study is presented, in which this method is shown together with the experimental analysis of the implemented algorithm.