The c.620 T > G mutation in rhodopsin found in the first mapped autosomal dominant retinitis pigmentosa (adRP) locus is associated with severe, early-onset RP. Intriguingly, another mutation affecting the same nucleotide (c.620 T > A) is related to a mild, late-onset RP. Assuming that both mutations are missense mutations (Met207Arg and Met207Lys) hampering the ligand-binding pocket, previous work addressed how they might differentially impair rhodopsin function. Here, we investigated the impact of both mutations at the mRNA and protein level in HEK293 cells and in the mouse retina. We show that, in contrast to c.620 T > A, c.620 T > G is a splicing mutation, which generates an exceptionally strong splice acceptor site (SAS) resulting in a 90 bp in-frame deletion and protein mislocalization in vitro and in vivo. Moreover, we identified the core element underlying the c.620 T > G SAS strength. Finally, we demonstrate that the c.620 T > G SAS is very flexible in branch point choice, which might explain its remarkable performance. Based on these results, we suggest that (i) point mutations should be routinely tested for mRNA splicing to avoid dispensable analysis of mutations on protein level, which do not naturally exist. (ii) Puzzling disease courses of mutations in other genes might also correlate with their effects on mRNA splicing. (iii) Flexibility in branch point choice might be another factor influencing the SAS strength. (iv) The core splice element identified in this study could be useful for biotechnological applications requiring effective SAS.