Soft sensors for regression applications (SSR) are inferential models that use online available sensors (e.g. temperature, pressure, flow rate, etc.) to predict quality variables which cannot be automatically measured at all, or can only be measured at high cost, sporadically, or with high delays (e.g. laboratory analysis). SSR are built using historical data of the process, usually provided from the supervisory control and data acquisition (SCADA) system or obtained from laboratory annotations/measurements. In the SSR development, there are many issues to deal with. The main issues are the treatment of missing data, outlier detection, selection of input variables, model training, validation, and SSR maintenance. In this work, a literature review on each of these topics will be performed, reviewing the most important works in these areas. Emphasis will be given to the methods and not to the applications.