In　the　work　of　legal　judgment　prediction,　the　most　common　tasks　are　crime　prediction　and　related　law　prediction.　These　works　can　be　regarded　as　a　multi-label　text　classification　task.　In　order　to　further　improve　the　accuracy　of　prediction　work,　this　paper　proposes　a　multi-task　deep　neural　network　classification　model,　named　M-AttBLSTM-CNN,　which　based　on　integrating　TextCNN　with　Att-BLSTM　model　to　achieve　high-precision　prediction　for　crime　prediction　and　related　law　prediction.　Automatically　extracting　more　rich　features　is　the　biggest　feature　of　this　model,　focusing　on　local　features　while　taking　the　full-text　information　into　account　at　the　same　time.　Firstly,　according　to　the　characteristics　of　the　two　sub-models,　We　combining　the　two　sub-models　in　parallel　to　obtain　more　feature　information.　Secondly,　the　bidirectional　LSTM　and　attention　mechanism　are　also　introduced　in　the　model　of　this　paper,　which　effectively　alleviates　the　model　over-fitting　problem　and　further　optimizes　the　model　feature　selection.　Finally,　the　experiments　are　carried　out　on　a　judicial　documents　data　set,　which　enjoys　a　large　number　of　corpora　up　to　626,600.　From　experiments,　the　proposed　model　has　better　performance　than　the　state-of-art　text　classification　models　including　SVM-TFIDF,　hierarchical　attention　network(HAN)　and　deep　pyramid　convolutional　neural　network　(DPCNN).　©　2019　IEEE.