An algorithm for estimating a continuous-time transfer function model directly from noise-free step response data was presented in Cluett and Wang (Chem. Engng Sci. 46, 2065-2077, 1991). The algorithm was based on the approximation of the process impulse response by a series expansion using the set of Laguerre functions (Lee, Statistical Theory of Communication, Wiley, 1960). This paper provides the analysis for the case when the algorithm is applied to noisy step response data. The bias and variance of the estimated model parameters are analysed and a frequency domain bound on the modelling error is estimated. Some direction is given in terms of operating conditions under which this algorithm can be expected to provide a good model, and also some practical suggestions are provided for pre-treatment of the noisy step response data which can greatly improve the model quality.