Handwritten　numeral　recognition　is　about　identifying　Arabic　numerals　and　various　coding　and　statistical　data　which　are　composed　by　a　few　of　special　symbols.　Feature　extraction　for　improved　recognition　rate　has　a　great　influence,　and　effective　classification　can　be　characterized　in　that　the　feature　is　separability,　reliability　and　independence.　Characteristic　dimension　should　as　little　as　possible.　To　meet　this　requirement,　classifier　needs　to　combine　various　features　to　put　them　together.　This　article　describes　the　extraction　method　of　handwritten　digits　eigen　values,　based　on　the　researches　on　features　of　several　handwritten　digits,　discussing　6　kinds　of　features,　they　are　Fourier　switch　features,　stroke　density　features,　contour　features,　projection　features,　the　barycenter　and　barycenter　distance　feature,　wide　grid　feature.　Finally　using　inner　and　outer　analogy　method　to　select　and　filter　the　features.　©　Springer-Verlag　Berlin　Heidelberg　2013.