The visual quality of urban streets is of vital importance for establishing a satisfying and comfortable experience for the residents in an urban community. It also has positive effects on urban vibrancy, public health, and social connections in that community. Numerous studies have been conducted to evaluate the visual quality at the urban street level using street view images. However, the spatial distribution of fine-grained visual quality inside an urban street is rarely investigated. This study presents a new approach for the evaluation of visual quality inside urban streets using mobile LiDAR point clouds. The semantic information of urban streets was first extracted from mobile LiDAR point clouds with a Gradient Boosting classifier. After that, seven well-known key design elements, including the green space factor, sky view factor, enclosure rate, volume index, vehicle occurrence rate, motorization rate, and diversity, were calculated from the classified point clouds using a three-dimensional (3D) visibility model. Finally, the visual quality at 1 m grid resolution inside urban street was achieved automatically by using a random forest model which was trained based on perception samples. This approach has been validated on two study areas and the results indicated that the proposed approach is able to quantitatively examine the visual quality difference inside urban streets. The results generated by the proposed method also match well with the common sense of urban design experts, which are useful for architects and designers to develop best practices in the urban micro-renewal project and to refine the urban planning processes.