3D data compression techniques can be used to determine the natural basis of radial eigenmodes that encode the maximum amount of information in a tomographic large-scale structure survey. We explore the potential of the Karhunen-Loève decomposition in reducing the dimensionality of the data vector for cosmic shear measurements, and apply it to the final data from the CFHTLenS survey. We find that practically all of the cosmological information can be encoded in one single radial eigenmode, from which we are able to reproduce compatible constraints with those found in the fiducial tomographic analysis (done with 7 redshift bins) with a factor of ~30 fewer datapoints. This simplifies the problem of computing the two-point function covariance matrix from mock catalogues by the same factor, or by a factor of ~800 for an analytical covariance. The resulting set of radial eigenfunctions is close to l-independent, and therefore they can be used as redshift-dependent galaxy weights. This simplifies the application of the Karhunen-Loève decomposition to real-space and Fourier-space data, and allows one to explore the effective radial window function of the principal eigenmodes as well as the associated shear maps in order to identify potential systematics. We also apply the method to extended parameter spaces and verify that additional information may be gained by including a second mode to break parameter degeneracies. The data and analysis code are publicly available at https://github.com/emiliobellini/kl_sample.