A particular aspect of uncertainty management in degradation-based reliability models is investigated. The presence of unobservable failure mechanisms in the reliability model is considered to be a source of variability greatly affecting the model predictive accuracy. A methodology is proposed to reduce the uncertainty effects imposed by the degradation data obtained in the presence of unobservable failure mechanisms. The key idea of the methodology is to transform the available degradation data so that variability of the critical degradation indicator values would be minimal. A numerical example utilizing degradation data is presented. The practical benefits of using the transformed degradation data are outlined.