The bacterial CRISPR/Cas9 system allows sequence-specific gene editing in many organisms and holds promise as a tool to generate models of human diseases, for example, in human pluripotent stem cells1,2. CRISPR/Cas9 introduces targeted double-stranded breaks (DSBs) with high efficiency, which are typically repaired by non-homologous end-joining (NHEJ) resulting in nonspecific insertions, deletions or other mutations (indels)2. DSBs may also be repaired by homology-directed repair (HDR)1,2using a DNA repair template, such as an introduced single-stranded oligo DNA nucleotide (ssODN), allowing knock-in of specific mutations3. Although CRISPR/Cas9 is used extensively to engineer gene knockouts through NHEJ, editing by HDR remains inefficient3,4,5,6,7,8and can be corrupted by additional indels9, preventing its widespread use for modelling genetic disorders through introducing disease-associated mutations. Furthermore, targeted mutational knock-in at single alleles to model diseases caused by heterozygous mutations has not been reported. Here we describe a CRISPR/Cas9-based genome-editing framework that allows selective introduction of mono- and bi-allelic sequence changes with high efficiency and accuracy. We show that HDR accuracy is increased dramatically by incorporating silent CRISPR/Cas-blocking mutations along with pathogenic mutations, and establish a method termed ‘CORRECT’ for scarless genome editing. By characterizing and exploiting a stereotyped inverse relationship between a mutation's incorporation rate and its distance to the DSB, we achieve predictable control of zygosity. Homozygous introduction requires a guide RNA targeting close to the intended mutation, whereas heterozygous introduction can be accomplished by distance-dependent suboptimal mutation incorporation or by use of mixed repair templates. Using this approach, we generated human induced pluripotent stem cells with heterozygous and homozygous dominant early onset Alzheimer's disease-causing mutations in amyloid precursor protein (APPSwe)10and presenilin 1 (PSEN1M146V)11and derived cortical neurons, which displayed genotype-dependent disease-associated phenotypes. Our findings enable efficient introduction of specific sequence changes with CRISPR/Cas9, facilitating study of human disease.