A new randomized response procedure is proposed based on Mangat and Singh two-stage randomized response model and keeps the respondents’ confidentiality. The basic idea of the suggested procedure is to increase the respondent confidentiality and to make him feel more protective, since his response is to be given after using three questions in the first stage not two questions, as Mangat and Singh model and Kim and Elam model suggested in their designs. The proposed model is extended to the stratified sample. The optimal sample size and the optimal allocation for the stratified random sample in the case of minimizing the variances of the estimators, are derived. An application for the suggested model to estimate the proportion of informal marriage among the students in the university level is given