Biological connectivity studies are crucial for explaining the structure of marine populations of species with a pelagic larval stage. Numerical modelling is a powerful tool for evaluating marine dispersal pathways of planktonic organisms. The high-resolution numerical circulation model HYCOM (HYbrid Coordinate Ocean Model) has been used to examine fish connectivity patterns in the Gulf of Mexico (GoM), but without biological validation. We examined the connectivity of fish larvae caught in the northwestern GoM by coupling ichthyoplankton surveys with numerical modelling and particle backtracking experiments. Fish larvae were collected with 200 m oblique bongo tows along two parallel transects extending from the edge of the shelf to deep waters of the Perdido region in the northwestern GoM (24°N to 26°N and 94.5°W to 97°W) during four cruises (June and October 2016, April and November 2017). Larvae of coastal and shelf-spawning species were used to infer offshore transport. In order to explore their dispersal pathways, the real-time HYCOM 1/25° model hourly output with the Navy Coupled Ocean Data Assimilation (NCODA) was used to simulate Lagrangian trajectories for each cruise and transect. Particles were seeded at sampling stations and the circulation model was run backward in time to infer larval origin. Patterns of spatial distribution and abundance of the coastal and shelf larvae caught in oceanic stations among cruises exhibited a reasonable agreement with the results of modelling exercises, and indicate the shelves of Tamaulipas and Texas were the main source of larvae to Perdido’s deepwater region. Our results suggest the combined use of ichthyoplankton surveys and ocean circulation models can yield insight into the dispersal pathways of larvae of neritic fish species to deep waters regions in the GoM. Likewise, these results imply that fish larval distributions are a useful tool for evaluating the reliability of using 2-D velocity fields from circulation models to infer larval transport at time scales of several weeks.