The estimation of models with time-varying coefficients is usually performed by Kalman-Bucy filtering. The two-sided filter proposed by Schlicht (1988) is statistically and computationally superior to the one-sided Kalman-Bucy filter. This paper describes the estimation procedure and the program package that implements the two-sided filter. Keywords: Kalman filtering, Kalman-Bucy, random walk, time-varying coefficients