This contribution seeks to compare the efficiency of several soft-sensor design methods using the datasets from three different case studies, including two industrial examples. The selected set of design methods compared considers different principles to design soft sensors, consisting of three consequent stages: (a) data preprocessing with input domain analysis, (b) input structure selection, and (c) model training. The input domain analysis explores the potential of various nonlinear structures within the input dataset. The results obtained indicate that multivariate feature selection approaches have the highest efficiency. Moreover, the nonlinear soft sensors achieve higher accuracy compared to linear ones.