Why the reading is wrong
A refractometer measures how much a liquid bends light and reports that as Brix, assuming the only thing dissolved is sugar. In finished beer a good deal of what is dissolved is ethanol, which has a much stronger refractive effect per gram. The instrument cannot tell them apart, so it reports a gravity that is too high — often dramatically so.
Before fermentation the reading is trustworthy. After it starts, it needs correcting.
The wort correction factor
Refractometers are calibrated against sucrose solutions. Wort is not sucrose, so even an unfermented reading runs slightly high. Dividing by a wort correction factor fixes this.
Typical values are 1.02 to 1.06 and 1.04 is a reasonable default, but yours is a property of your instrument and should be measured. Take a hydrometer reading and a refractometer reading of the same unfermented wort, convert the hydrometer reading to Brix, and divide the refractometer figure by it. Use that number from then on.
Terrill's models
Sean Terrill fitted these to measured beer data. Both take the corrected original and final Brix and return specific gravity.
The cubic is generally the better fit. Both are most reliable near terminal gravity, above roughly 60% apparent attenuation, and both degrade badly early in fermentation. If you want a mid-fermentation number, use a hydrometer.
What a refractometer is actually good for
It needs two drops instead of a cylinder full of beer, so it is unbeatable for checking pre-boil gravity, tracking the runnings during sparge, and confirming that fermentation has stopped moving. It is not the instrument for the final, definitive gravity reading — for that, pull a sample and float a hydrometer.