We propose a methodology that separates critical dimension (CD) variation depending on local pattern density into error sources; CD error on exposure mask, mid-range flare of exposure tool, and acid evaporation during post exposure baking (PEB). This methodology consists of two particular processes. One is over-coating process onto resist film before exposure, and the other is double exposure process using test reticles. The test reticles have line-and-space (L/S) array region and peripheral region that contains opaque area to avoid overlapping with L/S array region in double exposure process. Over-coating process allows separation of acid evaporation from other error sources. On the other hand, double exposure process enables elimination of CD error on exposure mask depending on local pattern density since double exposure process can make various pattern densities around fixed L/S mask pattern. In fact, the experimental CD results could provide the good agreement with estimated CD results using mid-range flare model. It has become clear that the influence of local pattern density increases with design rule shrinkage. Furthermore, the methodology revealed that the influence of each error source is greatly dependent on local pattern density. Consequently, the methodology is effective to separate CD variation on local pattern density into error sources.