Targeting observation is an observation strategy which can efficiently and effectively improve the quality of observations. Identification of sensitive regions is a key point in targeting observations. Three typical methods including Singular Vectors (SVs), Ensemble Transform Kalman Filter (ETKF), and the Conditional Nonlinear Optimal Perturbation (CNOP) are applied in the Lorenz96 model in this paper. Their advantages and disadvantages are discussed, and the reason why the ETKF method’s efficiency is unstable is discussed. The results show: For forecast time within 312 hours, the CNOP method’s forecast accuracy promotions are the highest among three methods, and its sensitive regions are small. While the SVs method has good results in the 72hour forecast, but sharply descends after that, and generally invalid after 120 hours. The ETKF method is not as good as other methods in the 72 hour forecast; moreover, though the comparison to the randomly selected sensitive regions, due to the serial observation processing is used, which ignores the relation with observation data, the ETKF method cannot successfully find the sensitive regions, which has the maximum covariance of signals and so its promotions to forecast accuracy is limited. It indicates how to search sensitive regions optimally is the key to promote the efficiency of the ETKF method.