In the attribute-based cryptosystems, user's identity is extended as a set of attributes. In order to solve the access control problem caused by the change of users' attributes, attribute-based encryption (ABE) schemes with attribute revocation were proposed. However, there are some problems like high revocation cost or coarse-grained revocation in most of the existing ABE schemes. Besides, the attribute key escrow problem is serious, that is the attribute authority can impersonate any user to decrypt the ciphertexts since the user's attribute private key is generated by the attribute authority himself. In order to remedy the above mentioned problems, the study proposes a ciphertext-policy attribute-based scheme supporting fine-grained attribute direct revocation, whose formal definition and security model are also presented. In the proposal, user's attribute private key is generated by the system authority and multiple attribute authorities jointly, so that each attribute authority's privilege can be effectively limited. Furthermore, the proposal constructs an efficient re-encryption method based on the access tree, which, together with the attribute revocation list, can be used to realize fine-grained attribute direct revocation with low revocation cost. By the formal security proof, the proposal is proven to have the characteristics of indistinguish ability under the adaptive chosen cipher-text attack and can protect the system from being attacked by the incredible authority. Compared to the similar schemes, the proposal can achieve higher computation efficiency and finer-grained attribute direct revocation.