In polarimetry, the condition number defines the stability and noise sensitivity that apolarimeter can achieve, being a key factor to understand its performance. In this work,we numerically calculate the error obtained for Stokes polarimetry with different numberof measurements and optimization strategies. Using these results, we show that not alldefinitions and metrics used in the scientific literature to calculate the condition numbersare physically robust and metrologically significant, and propose the best ones for predictingthe error in Stokes polarimetry. The selected definitions imply norms based on the singularvalue decomposition of the inverse, or pseudo inverse, of the instrument matrix associatedto the measurement procedure.