Abstract In experimental design for functional molecules, materials and products, complex relationships exist between the various experimental parameters and the target physical and chemical properties. Regression analyses with experimental data are a useful way to understand these relationships. A carefully constructed regression model can be used to search for functional products effectively. However, although these products can be found in the extrapolation domains of the existing data, the predictive ability of the model tends to be low in regions where the data density is low, and new candidates where the predicted values of a property are unreliable will not achieve the desired property values. Therefore, to search for new candidates in appropriate extrapolation domains, we consider the probability that a new candidate will have the intended property values and the reliability of a predicted property value for this candidate. The probability is calculated from a predicted value and the estimated prediction error, and the reliability is based on the data density. The proposed method is applied to both simulation data and aqueous solubility data, and the efficiency of the method is demonstrated through data analyses.