Robust　neural-network　controller　of　propeller　thrust　is　proposed　for　autonomous　underwater　vehicles　(AUVs).　The　cascaded　plant　consists　of　the　dynamics　of　surge　motion　of　an　AUV,　that　of　the　propeller　axial　flow,　that　of　the　propeller　shaft　and　that　of　the　electrically-driven　circuit.　Uncertainties　including　modeling　errors　and　external　disturbances　are　taken　into　account　simultaneously.　A　hybrid　control　strategy　is　proposed　to　the　cascaded　system　with　uncertainties.　An　on-line　robust　neural-network　is　used　to　compensate　the　modeling　errors　while　L2-gain　design　is　used　to　suppress　the　external　disturbances.　By　backstepping　method,　the　terminated　control　input　to　the　thrust　system　is　obtained.　Design　of　the　controller　with　L2-gain　performance　observes　the　recursive　Lyapunov　function　method,　which　guarantee　the　uniformly　ultimately　bounded　stability　of　tracking　system.　Simulation　results　demonstrate　the　validity　of　controller　proposed.