We propose a new blind multiuser signal model and detection framework for solving the near-far problem in synchronous CDMA in this paper. Compared with existing blind detectors, the proposed framework requires a minimum number of previously received signals, which is about the number of interfering users, and no subspace separation or sequence estimation. Hence its computation complexity and detection delay are much reduced. Following this framework, several blind multiuser detectors are developed using least squares (LS) estimation, best least unbiased (BLU) estimation and minimum mean-square error (MMSE) estimation criteria and a recursively adaptive procedure is developed for further decreasing the complexity. All these can be easily extended for asynchronous CDMA. The near-far performance of this framework and the trade-off between the complexity and performance are discussed. Computer simulations are provided to demonstrate the performance of the proposed schemes too.