The explosive growth of video applications and services on mobile devices has made it important to assess video quality. In this paper, we propose a no-reference video quality assessment method for mobile videos. Based on the analysis on common mobile video impairments, three features (blockiness, blurriness and noise) were extracted. The features are then trained to predict the DMOS (Differential Mean Opinion Score) through a support vector machine (SVM). To reduce complexity and increase adaptation, we capture a set of independent images from screen shot, and compute underlying features directly from the spatial domain. Dataset from a public database is used to train and test. Experimental results show that the proposed model provides satisfactory performance on characterizing the spatial domain impairments.