Abstract Study question What is the cause of loss of heterozygosity, frequently observed in embryos subjected to CRISPR-Cas9, an interhomolog-homologous recombination repair (IH-HR) or deficiency in DNA repair? Summary answer Vast majority of LOH observed in human zygotes subjected to CRISPR-based editing is explained by deficiency in DNA repair prior to genome activation, not IH-HR. What is known already Genome editing (GE) at the preimplantation stage, intended to correct disease-causing mutations, is controversial due to safety and ethical concerns. It has been suggested that a high frequency of LOH seen in zygotes subjected to GE could be explained by resolution of the double-stranded DNA breaks (DSBs) created using CRISPR-Cas9 by interhomolog-homologous recombination repair (IH-HR), a process with potential to correct deleterious heterozygous mutations. However, such hypotheses are controversial and it is unclear whether embryos at this stage development can utilise IH-HR. CRISPR-Cas9 targeted sites are difficult to interrogate with traditional methods, risking a misinterpretation of editing outcomes. Study design, size, duration We hypothesised that the frequently observed LOH affecting the CRISPR-targeted sites in human zygotes is caused by the deficiency in DNA repair at this developmental stage, resulting in complex genotypes which traditional molecular methods have failed to detect, and not IH-HR which uses the second un-cleaved parental allele as a template for repair. DSBs were induced at three genomic sites in embryos microinjected at the time of fertilisation (n = 27) in this IRB-approved study. Participants/materials, setting, methods Research embryos were cultured for 60 hours, then disaggregated. Individual cells underwent an array of genetic tests, including NGS, allowing determination of the repair pathway utilised and revealing any cytogenetic abnormalities. A novel state-of-the-art analytical workflow was developed to interrogate 25kb around CRISPR cleavage sites. This approach provided a means of detecting complex genomic rearrangements as well as unresolved DNA damage, which would appear as LOH when employing traditional PCR-based approaches for assessment. Main results and the role of chance Induction of DSBs using CRISPR-Cas9 was 100% efficient when applied at the time of fertilisation. LOH was confirmed to be common at targeted sites, affecting 69/216 (32%) loci examined in 108 blastomeres. Remarkably, the great majority of LOH events (>90%) were a direct consequence of inappropriate/failed DNA repair (unresolved DNA breaks or complex genomic rearrangements). Strikingly, not a single instance of IH-HR was observed when CRISPR reagents were introduced at fertilisation. This may be explained by the physical separation of the two parental genomes in distinct pronuclei at that time. Allele-drop out (ADO), a technical problem, accounted for <10% of all LOH events. In total, 46% of DSBs failed to undergo appropriate repair, resulting in segmental aneuploidy (21%), large-scale deletions (10%), inversions/translocations (10%), or loss of the targeted chromosome (5%). 45% of DSBs were resolved using error-prone non-homologous end joining (NHEJ), while only 9% utilised the homology directed repair (HDR) pathway required for correction of mutations. Application of CRISPR-Cas9 to embryos on day-3 of development (after embryonic genome activation - EGA) was less efficient, only producing a DSB in 6/37 embryos (16%). However, the mode of repair differed markedly, 100% employing HDR (p = 0.0001). In 4 cases, this involved IH-HR. Limitations, reasons for caution This data provides strong evidence that LOH recorded in earlier studies was incorrectly attributed to IH-HR and was actually due to abnormal/failed repair. CRISPR appears more effective after EGA, with efficient DNA repair dominated by desirable HDR. However, a larger number of embryos is required to corroborate this finding. Wider implications of the findings The use of CRISPR-Cas9 at the time of fertilisation is frequently associated with problematic, unintended outcomes, rendering the technique clinically inapplicable in the current format. Careful study design which allows determination of all possible editing outcomes is essential and advances our understanding of the underlying mechanisms of DNA repair. Trial registration number not applicable