Electroencephalogram　(EEG)　measurement,　being　an　appropriate　approach　to　understanding　the　underlying　mechanisms　of　the　major　depressive　disorder　(MDD),　is　used　to　discriminate　between　depressive　and　normal　control.　With　the　advancement　of　deep　learning　methods,　many　studies　have　designed　deep　learning　models　to　improve　the　classification　accuracy　of　depression　discrimination.　However,　few　of　them　have　focused　on　designing　a　convolutional　filter　to　learn　features　according　to　EEG　activity　characteristics.　In　this　study,　a　novel　convolutional　neural　network　named　HybridEEGNet　that　is　composed　of　two　parallel　lines　is　proposed　to　learn　the　synchronous　and　regional　EEG　features,　and　further　differentiate　normal　controls　from　medicated　and　unmedicated　MDD　patients.　A　ten-fold　cross　validation　method　is　used　to　train　and　test　the　model.　The　results　show　that　HybridEEGNet　achieves　a　sensitivity　of　68.78%,　a　specificity　of　84.45%,　and　an　accuracy　of　79.08%　in　three-category　classification.　The　result　of　EEG　feature　analysis　indicates　that　the　differences　of　spatial　distributions　and　amplitude　ranges　in　the　alpha　rhythm　(especially　at　approximately　10　Hz)　among　three　categories　might　be　distinctive　attributes　for　depression　discrimination.