Strain gradients have been cast in the form of geometrically-necessary dislocations (GND) to relate the length-scale dependence of strength and to determine potential sites for failure initiation. The literature contains various different incompatibility measures, the main ones being: the total form (∇×Fp), the rate form for large displacements (∇×γ̇anaFp), and the slip gradient form (∇γ̇a). Here these different approaches are compared rigorously for the first time. Obtaining GND densities when using the total form is a rank-deficit linear problem, solved by singular value decomposition (SVD) known as the Least Squares Minimization (L2 method). Alternative methods to find GND densities such as Karush-Kuhn–Tucker (KKT) optimization are also investigated. Both L2 and KKT method predict unrealistic GND densities on inactive slip systems leading to excessive strain hardening; even for a single crystal single slip case. Therefore, the restriction of GNDs to the active slip systems by using a threshold based on the total slip is found to be essential. This proposed restriction reveals relatively consistent results for various single crystal single slip cases including: simple shear, uniaxial tension, and four-point bending. In addition, the small numerical differences in the slip leads to large discrepancies in the flow stress due to error accumulation, even for strain-gradient-free uniaxial tension, hence a threshold for the GND density increment (2×102 m−2) is used in all models to avoid formation of artificial GND densities. Finally, the proposed method is applied to GND density evolution for a grain inside a polycrystal aggregate having a complex stress state. The total forms, that use the curl of the plastic deformation gradient, with the active slip system restriction giving the lowest incompatibility errors suggest them to be the most reliable GND measures.