Proper　orthogonal　decomposition　(POD)　is　proposed　to　decompose　and　reconstruct　multidimensional　electroencephalo-graph　(EEG)　signal.　With　this　method,　EEG　signal　can　be　represented　by　orthonomal　basis,　eigenfuction　mode　and　main　coordinate　component.　The　eigenfuction　mode　is　deterministic　functions　of　spatial　information,　and　the　main　coordinate　component　is　the　random　function　of　real　time　information.　In　this　modal　decomposition　way,　EEG　signal　can　not　only　be　compressed　in　a　certain　least　squares　optimal　sense,　but　also　be　predicted　for　any　random　scalp　position.　©　2012　IEEE.