AbstractType VI CRISPR enzymes have recently been identified as programmable RNA-guided, RNA-targeting Cas proteins with nuclease activity that allow for specific and robust target gene knock-down without altering the genome. However, we currently lack information about optimal Cas13 guide RNA designs for high target RNA knock-down efficacy. To close this gap, we conducted four massively-parallel Cas13 screens targeting the mRNA of a destabilized green fluorescent protein (GFP) transgene and CD46, CD55 and CD71 cell surface proteins in human cells. In total, we measured the activity of 24,460 guide RNA including 6,469 perfect match guide RNAs and a diverse set of guide RNA variants and permutations with mismatches relative to the target sequences.We find that guide RNAs show high diversity in knock-down efficiency driven by crRNA-specific features as well as target site context. Moreover, while single mismatches generally reduce knock-down to a modest degree, we identify a critical region spanning spacer nucleotides 15 – 21 that is largely intolerant to target site mismatches. We developed a computational model to identify guide RNAs with high knock-down efficacy. We confirmed the model’s generalizability across a large number of endogenous target mRNAs and show that Cas13 can be used in forward genetic pooled CRISPR-screens to identify essential genes. Using this model, we provide a resource of optimized Cas13 guide RNAs to target all protein-coding transcripts in the human genome, enabling transcriptome-wide forward genetic screens.