A　maximum　distribution　reduction　is　meant　to　preserve　not　only　all　deterministic　information　with　respect　to　decision　attributes　but　also　the　largest　possible　decision　class　for　each　object　of　an　inconsistent　decision　table.　Hence,　it　is　useful　to　compute　this　type　of　reduction　when　mining　decision　tables　with　data　inconsistency.　This　paper　presents　a　novel　algorithm　for　finding　a　maximum　distribution　reduct　of　an　inconsistent　decision　table.　Two　functions　of　attribute　sets　are　introduced　to　characterize　a　maximum　distribution　reduct　in　a　new　and　simple　way　and　then　used　as　a　heuristic　in　the　algorithm　to　search　for　a　reduction.　Complexity　analysis　of　the　algorithm　is　also　presented.　As　an　application　example,　the　presented　algorithm　was　applied　to　mine　a　real　surgery　database　and　some　interesting　results　were　obtained.　©　Springer-Verlag　Berlin　Heidelberg　2006.