Based on the QR decomposition technique, estimation method based on an orthogonal projection is proposed for a class of linear mixed effects models with incomplete observations. Under regularity conditions, the proposed estimator for fixed effects is proved to be asymptotically normal distributed, and then the confidence intervals for the fixed effects are constructed. The proposed estimator for fixed effects is not affected by the random effects, and then is more effective and robust compared with existing estimation methods. Some simulations and a real data application are also conducted for further illustrating the performances of the proposed method.