Productivity and wages tend to be higher in cities. This is typically explained by agglomeration economies, which increase the returns associated with urban locations. Competing arguments of specialization and diversity undergird these claims. Empirical research has long sought to confirm the existence of agglomeration economies and to adjudicate between the models of Marshall, Arrow and Romer (MAR) that suggest the benefits of proximity are largely confined to individual industries, and the claims of Jacobs (1969) that such benefits derive from a general increase in the density of economic activity in a particular place and are shared by all occupants of that location. The primary goal of this paper is to identify the main sources of urban increasing returns, after Marshall (1920). A secondary goal is to examine the geographical distance across which externalities flow between businesses in the same industry. We bring to bear on these questions plant-level data organized in the form of a panel across the years 1989 and 1999. The panel data overcome selection bias resulting from unobserved plant-level heterogeneity that is constant over time. Plant-level production functions are estimated across the Canadian manufacturing sector as a whole and for five broad industry groups, each characterized by the nature of their output. Results provide strong support for Marshall's (1920) claims about the importance of buyer-supplier networks, labour market pooling and spillovers. The data show spillovers enhance plant productivity within industries rather than between them and that these spillovers tend to be more spatially extensive than previous studies have found.