We analyse three public cosmic shear surveys; the Kilo-Degree Survey (KiDS-450), the Dark Energy Survey (DES-SV) and the Canada France Hawaii Telescope Lensing Survey (CFHTLenS). Adopting the 'COSEBIs' statistic to cleanly and completely separate the lensing E-modes from the non-lensing B-modes, we detect B-modes in KiDS-450 and CFHTLenS at the level of ∼ 2.7σ. For DES-SV we detect B-modes at the level of 2.8σ in a non-tomographic analysis, increasing to a 5.5σ B-mode detection in a tomographic analysis. In order to understand the origin of these detected B-modes we measure the B-mode signature of a range of different simulated systematics including PSF leakage, random but correlated PSF modelling errors, camera-based additive shear bias and photometric redshift selection bias. We show that any correlation between photometric-noise and the relative orientation of the galaxy to the point-spread-function leads to an ellipticity selection bias in tomographic analyses. This work therefore introduces a new systematic for future lensing surveys to consider. We find that the B-modes in DES-SV appear similar to a superposition of the Bmode signatures from all of the systematics simulated. The KiDS-450 and CFHTLenS B-mode measurements show features that are consistent with a repeating additive shear bias.1 Contributions beyond the first-order Born approximation (Schneider et al. 1998) and source clustering (Schneider et al. 2002b) can produce insignificant levels of B-modes for the current generation of shear surveys.Article number, page 1 of 32Theoretically the 2PCFs can be calculated through their relation to the shear power spectrum, P γ ,ξ − (θ) = ∞ 0 d 2π J 4 ( θ) P γ ( ) , Article number, page 2 of 32 Article number, page 4 of 32