In　practical　applications,　because　of　some　reasons,　such　as　node　failure　or　link　failure,　data　loss　is　a　normal　occurrence　in　wireless　sensor　networks.　For　many　applications　on　which　data　has　a　great　impact,　it　is　important　to　recover　the　lost　data.　Because　of　the　sensory　data　has　spatial　correlation　and　temporal　correlation,　it　is　difficult　to　recover　the　lost　data.　Consider　this,　in　this　paper,　we　propose　a　novel　algorithm　based　on　the　Markov　Random　Field　to　reduce　the　data　error　ratio　during　the　data　recovery　process.　In　our　proposed　algorithm,　we　use　the　curved　face　recovery　to　utilize　the　spatial　correlation　in　the　sensory　data,　then　we　explore　the　temporal　correlation　between　data　in　wireless　sensor　networks　so　that　it　can　improve　the　accuracy　of　data　recovered.　A　large　amount　of　experiments　show　that,　compared　to　other　relevant　works,　the　data　recovery　algorithm　we　proposed　can　reduce　the　error　rate　of　the　fitting　data,　especially　data　lost　seriously　in　the　wireless　sensor　network.