For　wastewater　treatment　plants,　a　large　number　of　process　variables　are　demanded　to　monitor　the　operation　of　the　system.　Given　the　problem　that　some　key　water　quality　variables　are　difficult　to　get　in　real-time,　a　soft-sensor　technology　is　devised　to　get　the　value　of　these　variables.　For　a　soft-sensor　model,　choosing　the　appropriate　input　variables　will　have　a　great　impact　on　its　performance.　In　this　paper,　automatic　relevance　determination　(ARD)　method　which　based　on　Gaussian　process　regression　(GPR)　is　proposed　to　select　the　appropriate　input　variables.　The　ARD　method　considers　the　nonlinear　mapping　relationship　from　input　variables　to　the　output　variable.　Moreover,　the　wastewater　treatment　plant　is　modeled　by　GPR,　which　requires　fewer　model　parameters　and　can　give　a　confidence　interval.　Finally,　an　example　of　the　wastewater　treatment　plant　is　used　to　prove　the　effectiveness　of　the　method.　©　2022　IEEE.