In this paper we apply the Indirect Inference method to estimate the parameters of a semi-Markov transition model when the data are subject to a complex form of censoring. There is no explicit expression for the likelihood function, and therefore Maximum Likelihood estimation is computationally burdensome. The econometric methodology of Indirect Inference is first tested on simulated data under various assumptions about the distribution of spell durations and transitions. Then, it is applied to labour market transitions between self-employment, wage-work, and unemployment using the 1986-1988 French labour force survey. Although the analysis is basically a reduced-form analysis, we motivate our transition model in terms of a structural search model.