Scheme for generating atomic entangled state is revisited. With the help of quantum nondemolition measurements, the present generation scheme can be succeeded in a nearly deterministic way. This improvement makes the present scheme more efficient than the schemes using nonunitary projective measurements, which are of probabilistic nature. Discussions show that our scheme is feasible within the current experimental technology. Entanglement was firstly introduced by Einstein, Podolsky, and Rosen and then was confirmed by Bell type experiment: the quantum correlations between entangled states violating Bell inequalities. Entanglement plays an central role in quantum information and quantum computation because of its intrinsic nonlocality [1]. The entanglement properties of quantum systems are presently attracting great attention in quantum information theory, such as quantum teleportation [2], quantum key distribution [3], quantum dense coding [4], etc. But quantum entanglement is generally sensitive to practical noise and technical imperfections, which will lead to the infamous decoherence effect. So many schemes with inherent robustness to diverse sources of noise have been proposed, such as the well known schemes for entangling atoms [5,6], macroscopic atomic ensembles [7,8], as well as for extensive experimental study [9,10]. However, many schemes to generate entanglement have utilized various tools, such as coherent control, feedback and imperfect measurements, which make the generation to be probabilistic.Recently, cross-Kerr nonlinearity was used to implement two-qubit CNOT gate [11] and Bell-state analysis (BSA) [12,13]. The idea of using weak cross-Kerr nonlinearities combined with strong coherent fields has been developed by several different authors and applied