We use observations of M3 to constrain thermohaline mixing. This is done by matching [C/Fe] and [N/Fe] along the RGB of M3. We find our models can explain observations if it is assumed there is a spread of ∼ 0.3 - 0.4 dex in [C/Fe] in the stars in M3 from their birth. We reproduce the full spread in [C/Fe] at the tip of the RGB. Thermohaline mixing can produce a significant change in [N/Fe] as a function of absolute magnitude on the RGB for initially CN-weak stars, but not for initially CN-strong stars, which have so much nitrogen to begin with that any deep mixing does not significantly affect the surface nitrogen composition.