We develop in this paper a generalization of the Indirect Inference (II) to semi-parametric settings and termed Semi-parametric Indirect Inference (SII). We introduce a new notion of Partial Encompassing which lays the emphasis on Pseudo True Values of Interest. The main difference with the older notion of encompassing is that some components of the pseudo-true value of interest associated with the structural parameters do correspond to true unknown values. This enables us to produce a theory of robust estimation despite mis-specifications in the structural model being used as a simulator. We also provide the asymptotic probability distributions of our SII estimators as well as Wald Encompassing Tests (WET) and advocate the use of Hausman type tests on the required assumptions for the consistency of the SII estimators. We illustrate our theory with examples based on semi-parametric stochastic volatility models.