Adaptive estimation of time-varying parameters in linearly parametrized systems is considered. The estimation time is divided into small intervals; in each interval the time-varying parameter is approximated by a time polynomial with unknown coefficients. A condition for resetting of the parameter estimate at the beginning of each interval is derived; the condition guarantees that the estimate of the time-varying parameter is continuous and also allows for the coefficients of the polynomial to be different in various time intervals. A modified version of the least-squares algorithm is provided to estimate the time-varying parameters. Stability of the proposed algorithm is shown and discussed. Simulation results on an example are given to validate the proposed method.