The issue of parameter identification arises whenever structural models are estimated. This paper develops a simple condition for local identification in linearized DSGE models. The condition is necessary and sufficient for identification with likelihood-based methods under normality, or with limited information methods that utilize only second moments of the data. Using the methodology developed in the paper researchers can answer, prior to estimation, the following questions: which parameters are locally identified and which are not; is the identification failure due to data limitations, such as a lack of observations for some variables, or is it intrinsic to the structure of the model.