Data mining is the process of extracting non-trivial patterns from large volumes of data. It generates insight and turns the data into valuable information. A critical yet common flaw when performing data mining is to disregard the geographic origin of the data. When this geospatial attribute is taken into consideration, the process is called geospatial data mining. This task essentially deals with the detection of spatial patterns in the data, the formulation of hypotheses and the assessment of descriptive or predictive spatial models. In this paper we present the results of spatial data mining on various education indicators of Bangladesh. At the greater district and zila level we consider literacy rates to be the primary education indicator. At the upazila level we look at primary school enrollment and average years of schooling for selected rural upazilas. We analyze the effect of educational establishments and poverty on these education indicators. We also compare the results of spatial regression model with classical regression model on this data. The results demonstrate that spatial lag model outperforms the classical model in different perspectives. We have found that education indicators have a tendency to produce spatial clusters. There is no obvious correlation between literacy rates and the presence of educational establishments but poverty does have a strong effect on education. It is clear that spatial data mining can provide interesting and useful insights for the government, economists and relevant decision makers. The results can also be used for causal analysis by domain experts.