In the last few decades, sampling theory has been given a substantial boost by the growing availability of geo-referenced finite populations. Unfortunately, geo-referentiation is often incomplete or affected by locational errors for a portion of the units. Spatial sampling methods produce efficient estimates but suffer from consequences of flaws in geo-referentiation. This paper proposes a mixed sampling strategy for finite populations where a portion of the units is not correctly geo-referenced. The strategy exploits the available spatial information in the sampling design and adopts traditional sampling techniques for the remaining part of the population. Statistical properties of the strategy are explained and studied through Monte Carlo experiments on simulated and real data. An analysis of results in terms of efficiency and optimal sample composition is performed. The design-based nature of the proposed approach and its adaptability to several practical situations make it a general and easy-to-implement tool, which can outperform pure spatial sampling designs in terms of efficiency in estimation.