The human hand is one of the primary biometric traits in person authentication. A hand image also includes a lot of attribute information such as gender, age, skin color, accessory, and etc. Most conventional methods for hand-based biometric recognition rely on one distinctive attribute like palmprint and fingerprint. The other attributes as gender, age, skin color and accessory known as soft biometrics are expected to help identify individuals but are rarely used for identification. This paper proposes an attribute estimation method using multi-convolutional neural network (CNN) from hand images. We specially design new multi-CNN architectures dedicated to estimating multiple attributes from hand images. We train and test our models using 11K Hands, which consists of more than 10, 000 images with 7 attributes and ID. The experimental results demonstrate that the proposed method exhibits the efficient performance on attribute estimation.