The spatial variability of physical properties, such as bulk density, penetration resistance and gravimetric moisture, obtained by applying geostatistics in precision agriculture, can effectively indicate the physical behavior of agricultural soils in longitudinal profiles. In this way, the spatial dependence of physical properties in streets of coffee plantations with different lengths was evaluated in the southern Minas Gerais, Brazil. For this purpose, five longitudinal profiles were measured in streets, each one with depths ranging from 0 to 0.60 m, in six layers of 0.10 m, being the database composed of 432 property, 144 by property, submitted to the ordinary kriging geostatistical method in order to obtain spatial variability maps using the R software. They were evaluated by the lower mean cross-validation error of theoretical models fitted by ordinary least squares (OLS), being detected in higher superficial layers, from 0 to 0.30 m, lower bulk density and lower penetration resistance, with variable gravimetric moisture in the length direction of some streets of coffee plantations, being that these properties presented different structures of spatial dependence for each street.