Soil-water characteristic curve (SWCC) is usually estimated by fitting a prescribed parametric model to a limited number of test data. Uncertainties in estimated SWCC are unavoidable due to a lack of test data, including uncertainty arising from selecting a parametric model and uncertainty in fitting parameters. This paper develops a reliability analysis approach for unsaturated slope under the Bayesian model averaging (BMA) framework. The proposed approach accounts, simultaneously, for SWCC parameter uncertainty and model selection uncertainty, and allows shedding light on their effects on unsaturated slope reliability analysis. The proposed approach is illustrated through an unsaturated cut slope example. Results show that, with a limited number of SWCC test data, explicit quantification of SWCC parameter uncertainty provides more insights into the variability of SWCC in comparison with only considering the best-fit of test data. More importantly, the model selection uncertainty might be considerably large when only a limited number of test data is available. In such a case, betting on a single model might result in an improper estimate of unsaturated slope failure probability. On the contrary, the proposed approach provides a model-independent estimate of unsaturated slope failure probability that is less sensitive to the selection of SWCC model among the pool of candidates.