Proteins with high stability are important in industrial processes and in biopharmaceuticals, hence the increased attention on protein engineering to optimize the stability of proteins. Prediction of protein stability is of high interest but still remains a challenging task. Several stability prediction methods have been made, but especially prediction of stabilizing protein variants has proven difficult to handle. Most single amino acid substitutions are destabilizing, and often it requires double or multiple amino acid substitutions to improve the stability of a protein. Here we describe a dual-reporter sensor system that allows simultaneous assessment of in vivo protein translation and protein folding without the use of direct fusion of reporter proteins. We show that the protein folding sensor is a good proxy for in vivo protein stability. By applying the dual-reporter system on mutant libraries, protein variants with altered protein stability can be obtained. We have compared experimentally determined stabilities of protein variants obtained using the dual-reporter system with computational stability predictors. The stability predictors in general show poor performance when compared to experimental values. We propose that the dual-reporter system can be used as a screening assay for protein stability to obtain protein variants with different stabilities that can be used to train computational stability predictors. We further describe a global analysis approach to accurately determine protein stability from temperature and solvent denaturation. We demonstrate that the free energy of unfolding, G, and the temperature midpoint, Tm, can be accurately determined.