With　the　rapid　development　of　the　Internet,　people　have　variety　in　their　life　style.　While　the　Internet　is　developing　rapidly,　some　malware　has　also　been　generated,　and　the　number　of　malicious　codes　grows　exponentially,　which　seriously　affects　the　security　of　the　Internet.　In　the　prevention　of　malicious　code,　the　accurate　classification　of　malicious　code　is　the　most　critical　components.　At　present,　researchers　mainly　focus　on　static　detection　and　dynamic　detection　to　study　malicious　code　variant　detection　schemes.　In　the　study　of　malicious　code　classification,　this　paper　first　analyzes　the　limitations　of　static　detection　and　dynamic　detection,　then　employs　convolution　neural　networks　to　classify　and　identify　malicious　code,　and　finally　implements　an　efficient　malicious　code　detection　system　based　on　convolutional　neural　networks.　Our　malware　detection　model　realizes　the　idea　of　automatic　detection　of　malicious　code　variants.　Experimental　results　demonstrated　that　the　proposed　malicious　code　detection　system　is　efficient.