This paper describes how to evaluate the accuracy of prognostic estimates for remaining-useful life (RUL), state of health (SoH), and prognostic horizon (PH). The primary goal of prognostics is to accurately predict future failure in systems. Condition-based maintenance (CBM) systems use condition-based data (CBD) to detect degradation and project that degradation to a failing level at a future time. Prognostic estimates are produced by a Prognostic Health Monitoring (PHM) system. The process comprises three stages: sensing, data processing, and prediction. The time when data is sampled, the amplitude of features extracted from that data, the rate at which the amplitudes change, and a model for that data are used to estimate that future time of failure and to calculate prognostic estimates (RUL, SoH, and PH): collectively known as prognostic information. Evaluation of the accuracy of prognostic information is critical and necessary to install confidence in those estimates. Evaluation begins with knowing that ideal, zero-error estimates are not achievable in practice because of, for example, initial-estimate errors, offset errors due to sampling, noise, and data nonlinearity. Performance metrics to evaluate the accuracy of RUL, SoH, and PH estimates are developed and tabulated: convergence efficiency, prognostic distance (PD), and PH accuracy (PHα). Other prognostic terms, names, and descriptions are also provided and examples of evaluation methods and how to improve prognostic accuracy are included.