To　improve　the　performance　of　motion　estimation　(ME),　a　new　fast　motion　estimation　algorithm　is　proposed　for　motion　complexity　decision　(MC-FME).　The　algorithm　adaptively　estimates　the　frame-level　motion　complexity　by　using　the　frame　texture　information　and　the　macro-block　residual　value　at　first,　and　then　adaptively　estimates　the　macro-block-level　motion　complexity　by　calculating　the　variations　of　predictions　in　the　prediction　set　of　the　macro-block　motion　vectors.　The　complexity　is　finally　used　to　estimate　the　motion.　Experimental　results　and　comparisons　with　the　diamond　search　algorithm　and　the　hierarchical　motion　estimation　algorithm　show　that　the　proposed　algorithm　can　get　5%　to　21%　reduction　in　the　number　of　matched　points　and　save　3%　to　5%　motion　estimation　time　while　the　image　quality　is　guaranteed.