Nature reserves are considered effective means for protecting endangered and threatened species and critical ecosystems. Unfortunately, resources devoted to nature reserves are scarce, therefore establishing an effective conservation plan is an economic issue as much as it is ecological. Spatial attributes have been found important to affect the protected species' migration patterns, genetic exchanges, outside disturbances, etc. To promote the survival chances of protected species, spatial coherence of the reserve should be taken into account when choosing reserve sites. This raises two important issues: (1) what spatial attributes a reserve should possess to be an effective reserve, and (2) how can we select a subset from a large number of potential habitat sites to optimally allocate limited conservation resources. Although existing studies in the biological conservation literature do not provide explicit information about specific effects of different spatial attributes on protected species, generally large, compact, and contiguous reserve configurations are believed to work better. In the last decade, especially in recent years, many linear and nonlinear models were presented in the literature to incorporate various spatial attributes of the selected sites and their computational efficiency was demonstrated using empirical or hypothetic data. This paper summarizes the recent literature in this area with a particular focus on the modeling methods that addressed spatial attributes of the selected sites using linear integer programming. Four major spatial attributes are considered: (1) connectivity, (2) fragmentation and distance between selected sites, (3) boundary length and compactness, and (4) reserve area and core and buffer zones. Earlier studies have addressed these issues using rule-based heuristic algorithms. However, it has been shown in the literature that this method is inefficient in terms of optimal utilization of conservation resources because solutions obtained from heuristic algorithms often deviate significantly from the true optimal solutions. The spatial attributes mentioned above can also be modeled using linear integer programming that involves yes/no type decision variables, and small to medium scale reserve design models can be solved conveniently using off the shelf optimization software to find exact optimal solutions, and therefore to guide the optimal utilization of limited resources. Unfortunately, currently available optimization software have limited capability in solving large size site selection problems and their computational efficiency may be a bottleneck in practice. Furthermore, most of the models that incorporate spatial attributes are deterministic in the sense that they do not consider the uncertainty in species' occurrence in potential reserve sites or the probability of urban development for potential reserve sites. Some other models incorporated both types of uncertainty but they excluded spatial attributes of the selected sites. Incorporating both uncertainty and spatial attributes emerges as a new and important research direction in this area. The paper also elaborates on potential research topics in biodiversity and ecosystem conservation and related problems in China. Inefficient data, both on the species presence in potential sites and the cost-benefit values associated with individual sites may become an obstacle for applications of site selection models in real conservation practices in China.