Long-term, trouble-free operation of ventricular assist devices (VADs) is critical to the patient. A catastrophic failure of the VAD could cost the patient's life, thus defeating the purpose of the device. The targeted 90% 5 year reliability also implies that the average device life would exceed the 5 year limit. Time based explantation of the device after the fifth year will replace many devices with significant additional life, subject the patient to unnecessary surgical risk, and increase costs. To preclude the need for time based replacements and prevent catastrophic failures, a condition monitor is proposed in this article for early detection of faults in VADs. To develop this monitor, the effectiveness of various sensing and monitoring methods for determining the VAD condition is investigated. A Hemadyne pump was instrumented with a set of eight sensors, and a series of experiments were performed to record and analyze signals from the normal and abnormal pumps with five different faults. Statistical, spectral, envelope, and ensemble averaging analyses were performed to characterize changes in sensor signals due to faults. Experimental results indicate that statistical and frequency information from the acceleration and dynamic pressure signals can clearly detect and identify various VAD faults.