Evaluation of reliability performance in every power system has to be done within a cost-benefit framework. This approach, however, is a very time consuming task, especially for systems that contain a large number of possible configurations, so simpler techniques referred to the calculation of reliability indices are used. In small autonomous power systems (SAPSs), such an evaluation uses mainly deterministic criteria. This approach, however, cannot be applied in SAPS that contain only renewable energy sources, due to the intermittent nature of the provided energy. In this paper, a complete reliability cost and worth analysis is implemented for these systems, combined with the calculation of some basic probabilistic indices, in order to discover their performance and propose the appropriate of them as a criterion of optimal system configuration. This paper proposes that normalized energy reliability indices as system minutes and energy index of unavailability can be used as adequate criteria of system's optimal performance. This conclusion is validated through a large number of sensitivity analysis studies that are based on different maximum annual loads and different mix of load types. (c) 2008 Elsevier Ltd. All rights reserved.