It was recognized recently that uptake characteristics of PET images could be used for planning and adapting radiotherapy treatment based on predicted outcome risks of individual patients. In this work, we are investigating PET uptake features as prognostic factors for patients with nonsmall cell lung cancer (NSCLC). Methods based on intensity-volume histogram (IVH) and extracted morphological features from regions of interest (ROI) such as shape deformations and texture heterogeneity are evaluated. Seventeen NSCLC patients who received 3D-CRT are analyzed. Sixteen patients with pre-radiotherapy FDG-PET were analyzed for local failure using the gross tumor volume (GTV) as ROI. Nine patients with post-radiotherapy FDG-PET were analyzed for pneumonitis. The lung minus GTV was selected as the ROI in this case. About thirty candidate variables were extracted from each case, which included: ROI volume, SUV descriptors, total lesion glycolosis (TLG), IVH variables, and local texture variability metrics. Model building approaches based on logistic regression and machine learning were evaluated and corresponding Spearman's rank correlation (rs) was reported. Our preliminary results for local failure indicate that GTV and TLG had the highest correlation (rs= 0.476, p=0.031) while meanSUV, maxSUV, and local texture homogeneity showed modest association. A combined logistic model of TLG and V90 yielded rs=0.616 (p=0.009). This is slightly improved using a quadratic kernel (rs=0.644, p=0.006). In pneumonitis analysis, local contrast and homogeneity had the highest correlation with rs=0.725 and rs=-0.725 (p=0.014). These results imply intensity-volume effect in predicting local failure and significant local heterogeneity association in predicting onset of pneumonitis.