To combat severe soil erosion and improve the land productivity, construction of protection forests has been widely implemented as one of the key ecological restoration efforts on the Loess Plateau in Northwestern China. After giving a brief introduction of the concept of protection forests defined by the various researchers, several issues with respect to the goal of protection forests, the scales and the principles for developing protection forests were examined from the viewpoints of restoration ecology and eco-hydrology. A few recommendations were proposed to provide the theoretical foundations for construction of protection forests. We argue that, in addition to soil erosion control, the ultimate of protection forests in the Loess area should serve both water regulation and erosion control. We should recognize the hierarchical system of protection forests. In addition to the structural-optimization techniques at the stand level and the spatial arrangements measures at the watershed level, determination of both zonal vegetation types and the coverage on the regional scale should be considered in designing protection forests. To succeed in constructing protection forests and fulfilling the established targets, several theories should be followed in practice, including the principle of water balance, the theory of ecological suitability, the biological diversity, and the theory of ecosystem regeneration and succession. Construction of protection forests should not alter the balance of water availability and tree growth requirement. Dominant tree species of secondary forests should be encouraged to enhance forest succession and bio-diversity when implementing the conventional principle of “tree and site match”. We conclude that the following four key aspects should be reinforced: (1) evaluation of the vulnerability of protection forests, and developing adaptation strategies for managing protection forests under a changing climate and variability; (2) implementation of the principle of “tree-site match” by assessment of suitability of reforestation at high spatial and temporal resolution based on the interactive relations between soil, hydrology, vegetation, and climate; (3) predicting the spatial distribution of potential vegetation of the loess area using restoration ecology theories and constructing the protection forests with the pattern of close-to-nature to restore the vegetation of the Loess area; (4) understanding the hydrological responses of protection forests to the change in vegetation patterns, species composition, as well as coverage using physically-based distributed eco-hydrological models; developing theories for forest restoration design on forest structure and spatial distributions of protection forests.