This paper proposes a state variable filter approach to continuous time system identification. Two topics are studied in the paper. The first topic is related to the choice of state variable filters. The strategy we adopt is to adjust the time constants of the state variable filters so that a prediction error criterion is minimized. As a result, the estimated model reaches a balance between bias and variances shown by a simulation example. The second topic is related to the choice of model structure. We extend a multiple model estimation algorithm, developed using UD factorization, to continuous time sysem identification. The estimation algorithm generates a set of candidate models, among which the 'best' model structure is found. A simulation example is used to demonstrate the efficacy of the proposed procedure, and an industrial case study on a food cooking extrusion process is given to illustrate the applicability of the algorithm.