Visualization of large spatial data is a complex process and many GIS packages use their own assumptions or special storage formats to render quality data. Im- age pyramids and tile based rendering are used for raster data visualization but vector dataset uses feature limiting mechanism for progressive rendering. For vec- tor dataset these methods improve rendering but have very little visual appeal and make visual interpretation difficult. This paper proposes a method for rendering Vector data, called Vector Pyramids, that is designed to improve rendering and visual interpretation of large vector datasets. This method is a combination of ge- ometric aggregation and attribute level amalgamation to generate multi-scale data that can be appropriately called for rendering based on both the zoom level and map display extent. It is an automatic method, that mines the attributes of the the- matic layer data itself to suitably amalgamate and hierarchically rank them. For a dataset with more than 200,000 features, the results of the algorithm show that Vec- tor pyramids perform at orders of 2 to 14 times faster than current implementations of vector data rendering, depending on the zoom level and map display extent. This algorithm is suitable for both desktop or web-based rendering of geospatial data.