The geography of fuel prices has many various implications, from its significant impact on accessibility to being an indicator of territorial equity and transportation policy. In this paper, we study the spatio-temporal patterns of fuel price in the US at a very high resolution using a newly constructed dataset collecting daily oil prices for two months, on a significant proportion of US gas facilities. These data have been collected using a specifically-designed large scale data crawling technology that we describe. We study the influence of socioeconomic variables, by using complementary methods: Geographically Weighted Regression to take into account spatial non-stationarity, and linear econometric modeling to condition at the state and test county level characteristics. The former yields an optimal spatial range roughly corresponding to stationarity scale, and significant influence of variables such as median income or wage per job, with a non-simple spatial behavior that confirms the importance of geographical particularities. On the other hand, multi-level modeling reveals a strong state fixed effect, while county specific characteristics still have significant impact. Through the combination of such methods, we unveil the superposition of a governance process with a local socio-economical spatial process. We discuss one important application that is the elaboration of locally parametrized car-regulation policies.