The implementation into service of accelerometric health monitoring systems of mechanical power drives on helicopters has shown that the generation of false failure alarms is a critical issue. The paper presents a combined application of several multivariate statistical techniques and shows how a monitoring method which integrates these tools can be successfully exploited in order to improve the reliability of the diagnostic systems. The first phase of the research activity was addressed to exploring the potential advantages of using multivariate classification /discrimination/anomaly detection methods on real world accelerometric condition monitoring data. The second phase consisted of an implementation into actual service of an innovative integrated multivariate health monitoring system based on a third-level multivariate processing of the condition indicators. A monitoring method which integrates several multivariate statistical techniques was developed and implemented in an efficient integrated tool. When applied to actual data collected on several helicopters in service, this method proved to be able to distinguish with very high level of statistical confidence true failure situations from false anomaly alerts that had been indicated as failures by other health monitoring systems.