Abstract: In recent years, the targeting observations techniques are becoming more crucial for improving forecast skill in the numerical weather predicition. However, how to qualify the effect of applying adaptive observation on forecast errors deserves further discussion. Singular vectors method was used to analyze the sensitivity of forecast error in a blizzard storm event happening in 4 March 2007 which struck the northeastern China severely. Under constraining of energy norm defined by kinetic and potential energy, the singular vectors were calculated and their accuracy was verified, and the threedimensional structure of singular vectors was studied in detail in this paper. The sensitivity area was defined using the first several singular vectors. The pseudoinverse initial perturbation was used as an analysis error to test the sensitivity of forecast errors in the verificational region to the observations in the different regions. The results showed that when sampling extra observations in small regions, it was possible to improve the level of forecast only by improving the analysis error in the sensitivity area. Adding observations in the nonsensitivity areas had little impact on the forecast improvement in the validation area,suggesting that taking targeting observations in the sensitivity area defined by singular vectors is able to use finite observations and resources more efficiently for reducing the forecast error, and thus enhancing the accuracy of forecast.