Feature recommendation is one of the most critical and challenging problems in modern digital intelligence system. However, it is difficult to ensure privacy protection in many situations. To address this challenge, a deep federated feature recommendation method, called DF_Rec, is designed. Particularly, deep decentralized federated aggregation learning (DFAL) is jointly developed based on the ingenious combination of several deep frameworks and federated aggregation schemes. Extensive experiments are performed on three authoritative datasets, demonstrating that DF_Rec outperforms existing outstanding systems significantly.