A method for performing software error detection and prediction. The method includes identifying a plurality of software components in a computer software product. For each of the software components of the plurality of software components, the risk-relevant historical data pertaining to the respective software component is measured, then classified into at least a set of risk-increasing data and a set of risk-decreasing data. The set of risk-increasing data and the set of risk-decreasing data are then normalized, and a failure risk value for the respective software component is calculated by subtracting a weighted sum of the normalized values for the risk-decreasing data from a weighted sum of the normalized values for the risk-increasing data.