A self-validating soft sensor is proposed that not only can perform self-diagnostics and self-reconstruction, but also generate a variety of output data types, including the prediction values, input sensors status of soft sensor and the uncertainty values which represent the credibility of soft sensor’s output. The input sensors are validated before performing a prediction by principal components analysis (PCA) model. These validated data are then employed for subsequent recursive partial least square (RPLS) prediction. Other than input sensor validation and modeling for prediction, a t-statistic confidence interval is created and the status of input sensors is offered. By using this self-validating soft sensor, we can determine the work condition of the soft sensor and take proper actions in real time. The usefulness of the proposed method is demonstrated through a case study of a wastewater treatment process.