Lung carcinoma is the most prevalent type of cancer, and it is responsible for more deaths than other types of cancer . Unlike other imaging modalities used for lung cancer diagnosis, such as magnetic resonance imaging (MRI) and computed tomography (CT), coherent anti-Stokes Raman scattering (CARS) offers submicron spatial resolutions and video-rate temporal resolution along with chemical specificity. This method potentially allows label-free differential diagnosis of lung cancer at high resolution. This study aims at demonstrating the feasibility of using CARS to differentiate normal, benign and different subtypes of lung cancers. Human lung lesions, including small cell carcinoma, adenocarcinoma, squamous cell carcinoma and intestinal fibrosis, were imaged ex vivo using a CARS microscope and compared to their corresponding hematoxylin and eosin (H&E) staining results. Our data shows that CARS is capable of identifying cellular features in a similar way as H&E stain, and these features could be used to characterize different types of lung lesions from each other. Therefore, the CARS technique offers a new strategy for differential diagnosis of lung lesions using pathologically prevalent criteria, and thus could be further developed for in vivo and real time diagnosis of early stage lung cancer.