Recently, visual attribute learning and usage have become a popular research topic of computer vision. In this work, we aim to explore which low-level features contribute to the modeling of the visual attributes the most. In this context, several low-level features that encode the color and shape information in various levels are explored and their contribution to the recognition of the attributes are evaluated experimentally. Experimental results demonstrate that, the colorSIFT features that encode local shape information together with color information and the LBP features that encode the local structure are both effective for visual attribute recognition.