We examined the diagnostic accuracy of the cumulative smoking dose for identifying the epidermal growth factor receptor (EGFR) exon 19 deletion and L858R mutation among Japanese patients with non-small-cell lung cancer (NSCLC). EGFR mutations in exon 19 and exon 21 were determined in 1001 NSCLC patients. A receiver-operating characteristic (ROC) curve methodology was applied to estimate the diagnostic accuracy. EGFR mutations were detected in 314 patients (31.4%). A cumulative smoking dose of less than 13 pack-years (PY) was the optimal cut-off point for predicting a positive EGFR mutation status, producing a balance between the sensitivity (73.5%) and the specificity (77%). The area under the ROC curve was 0.77, indicating that the smoking dose had a moderate diagnostic accuracy. The median survival time or the median progression-free survival time of patients who had smoked less than 13 pack-years (PY) were 18.6 and 6.3 months, respectively, while those of patients with equal to or more than 13 PY were 9.6 and 2.4 months, respectively. The overall survival (OS) and progression-free survival (PFS) rates were significantly different between the two groups (OS: hazard ratio [HR] = 0.64, 95% confidence interval [CI] = 0.51-0.80, P = 0.0001) (PFS: HR = 0.58, 95% CI = 0.47-0.71, P < 0.0001). Our study indicated that the smoking dose predicted EGFR mutations with a moderate diagnostic accuracy. Thus, patients who have smoked less than 13 PY might be candidates for gefitinib treatment when EGFR mutation status cannot be determined.