Soil apparent electrical conductivity (ECa) is an effective indicator of soil variability as- sociated with many attributes relevant to crop production, being widely employed for site-specific management. Because ECa is closely related to soil moisture, topographic features along the field might change ECa readings within short distances reducing its spatial dependence, and thus the accuracy of spatial prediction. Within this context, we tested a filtering process that removes ECa data located on terraces across the field. The method was evaluated and compared with the original dataset based on the ac- curacy of spatial prediction and clustering performance for management unit delinea- tion. The results demonstrated that filtering out ECa readings from terraces increases the performance of spatial prediction, with higher accuracy and lower residuals. These results reflected on the clustering process, where the filtered dataset yielded more re- liable management units, confirming the feasibility of the proposed method.