Condition monitoring of lithium ion batteries plays a critical role in the battery management system of a hybrid electric or full electric vehicle. Battery fault conditions such as overcharge and over discharge causes significant variations of parameters from nominal values and can be considered as separate models. In this paper, multiple- model adaptive estimation techniques have been successfully applied to fault detection and identification in lithium-ion batteries. The diagnostic performance of a battery depends greatly on the modeling technique used in representing the system and the associated faults under investigation. Here, both linear and non-linear battery modeling techniques are evaluated and the effects of battery model and noise estimation on the over-charge and over-discharge fault diagnosis performance are studied. Based on the experimental data obtained under the same fault scenarios, the non-linear model based detection method is found to perform much better in accurately detecting the faults in real time when compared to those using linear model based method.