Lung carcinoma is the most prevalent type of cancer in the world, and it is responsible for more deaths than other types of cancer. During diagnosis, a pathologist primarily aims to differentiate small cell carcinoma from non-small cell carcinoma on biopsy and cytology specimens, which is time consuming due to the time required for tissue processing and staining. To speed up the diagnostic process, we investigated the feasibility of using coherent anti-Stokes Raman scattering (CARS) microscopy as a label-free strategy to image lung lesions and differentiate subtypes of lung cancers. Different mouse lung cancer models were developed by injecting human lung cancer cell lines, including adenocarcinoma, squamous cell carcinoma, and small cell carcinoma, into lungs of the nude mice. CARS images were acquired from normal lung tissues and different subtypes of cancer lesions ex vivo using intrinsic contrasts from symmetric CH2 bonds. These images showed good correlation with the hematoxylin and eosin (H&E) stained sections from the same tissue samples with regard to cell size, density, and cell-cell distance. These features are routinely used in diagnosing lung lesions. Our results showed that the CARS technique is capable of providing a visualizable platform to differentiate different kinds of lung cancers using the same pathological features without histological staining and thus has the potential to serve as a more efficient examination tool for diagnostic pathology. In addition, incorporating with suitable fiber-optic probes would render the CARS technique as a promising approach for in vivo diagnosis of lung cancer.