Zebrafish have practical features that make them a useful model for higher-throughput tests of gene function using CRISPR/Cas9 editing to create ‘knockout’ models. Due to the large number of available tools to design CRISPR assays and diversity of theories/model systems they were originally built on, we sought to systematically compare computational and empirical approaches for predicting gene-editing efficacy in zebrafish. We subjected zebrafish embryos to CRISPR/Cas9 with 50 different guide RNAs (gRNAs) targeting 14 genes and assayed individual editing efficiencies. We compared our experimental in vivo efficiencies in mosaic G0 embryos with those predicted by seven commonly used gRNA design tools and found large discrepancies between methods. Assessing off-target mutations (predicted in silico and in vitro) found that the majority of tested loci had low in vivo frequencies (<1%). Moreover, understanding that recent segmental duplications in the zebrafish genome could exacerbate CRISPR targeting of individual genes, we cataloged these loci and have made them available as a resource. Lastly, we assessed the transcriptome of negative ‘mock’ control CRISPR larvae injected with Cas9 enzyme or mRNA with no gRNA using RNA-seq and identified differentially expressed genes with high variability between injections. Using these same data, we discovered on average ~60 putative somatic mosaic frameshift mutations impacting genes per pool of injected larvae, potentially due to background cutting of DNA with Cas9 in the absence of gRNA. To verify this previously unreported phenomenon in zebrafish, we validated seven of twelve genes tested carrying low frequency mosaic somatic mutations in the genomes of a separate batch of injected larvae. These results suggest that negative control embryos may carry mutations within genes leading to spurious phenotypes. Overall, our results provide a valuable resource for the zebrafish community for the design and execution of CRISPR/Cas9 experiments.