For the purpose of fault detection in a reformate supplied fuel cell system, an adaptive observer is proposed in this paper to estimate the carbon monoxide (CO) molar fraction within the inlet gas of the fuel cell. Though CO accounts only for a small fraction in the hydrogen rich reformate, when adsorbed on the anode catalyst layer, it can considerably reduce the cell voltage, which is dangerous for the system. A high CO rate is thus an important fault indicator, however, existing sensors for CO molar fraction measurement in a gas flow are not suited to on-board applications. It is then useful to design an algorithm for its estimation from available measurements. A reduced model based on electrochemical phenomena is first briefly described and validated with experimental data. Then, we present the adaptive observer estimating the CO molar fraction, associated with a robust fault detection criterion. Finally, the algorithm is validated on transient experiments.