Class II CRISPR-Cas RNA programmable DNA endonucleases enable high efficiency genome editing across the biological diversity for research, industrial, and biomedical applications. Human genome editing with CRISPR-Cas just recently made its debut in human clinical trials and has immense therapeutic potential to fix disease-causing mutations at the level of DNA. Ensuring the integrity and safety of research, industrial, and biomedical applications of CRISPR-Cas necessitates efficient, versatile, and comprehensive methods to evaluate of the specificity of genome editing. Here, we optimize the efficiency and characterize the targeting specificity of SpCas9 to ensure robust cleavage activity while minimizing off-target activity in human cells. We characterize SpCas9 mismatch tolerance between the guide RNA and target, and provide data-driven design software to guide the selection of high fidelity Cas9 targets. We find that SpCas9 binding activity is not predictive of DNA cleavage, limiting the efficacy of Cas9 ChIP for unbiased evaluation of Cas9 off-target activity. Alternatively, we demonstrate that insert capture – insertion of short DNA fragments at double strand breaks (DSBs) by non-homologous end-joining  (NHEJ)– provides unbiased genome-wide identification of off-target cleavage by Cas9 as well as relative rates of indel, chromosomal rearrangement, and translocation accompanying NHEJ repair. However, insert capture is largely limited to use in model cell lines and is fundamentally limited in sensitivity due to labeling of low frequency errors in DSB repair. To directly labeling DSBs from cell culture or tissue samples, we adapted BLESS (direct in situ breaks labeling, enrichment on streptavidin and next-generation sequencing) and BLISS (Breaks Labeling In Situ and Sequencing) for unbiased genome-wide analysis of CRISPR-Cas specificity. Finally, we consider how human genetic variation will affect the targeting specificity of CRISPR-Cas endonucleases for therapeutic applications. Using the ExAC and 1000 Genomes datasets we find that human variation has important implications for Cas enzyme choice as well as target efficacy and safety. From this analysis, we provide a framework for the design of CRISPR-based therapeutics to maximize efficacy and safety across patient populations.