Abstract: Motivated by a missing data problem from the centrifuge experiments for developing a new anti-icing nanocoating, we consider a semiparametric transformation model with normal error terms and with the response subject to selection bias. Similar to Heckman’s two-stage procedure, probit regression is used to obtain the pseudo-regressor in the first stage and then the pseudo-regressor is used in the linear regression of the second stage. In the second stage, both the unknown transformation function and the parameters of interest are to be estimated. We propose an iterative algorithm based on two unbiased estimating equations. The resulting estimators are both consistent and asymptotically normal. Simulation results show that the proposed method also works well under finite sample situations. The proposed method is illustrated using the motivating example and a famous dataset used in Econometrics.