We implement and investigate a method for measuring departures from scale-invariance , both scale-dependent as well as scale-free , in the primordial power spectrum of density perturbations using cosmic microwave background ( CMB ) C _ { \ell } data and a principal component analysis ( PCA ) technique . The primordial power spectrum is decomposed into a dominant scale-invariant Gaussian adiabatic component plus a series of orthonormal modes whose detailed form only depends the noise model for a particular CMB experiment . However , in general these modes are localised across wavenumbers with 0.01 < k < 0.2 { Mpc } ^ { -1 } , displaying rapid oscillations on scales corresponding the acoustic peaks where the sensitivity to primordial power spectrum is greatest . The performance of this method is assessed using simulated data for the Planck satellite , and the full cosmological plus power spectrum parameter space is integrated out using Markov Chain Monte Carlo . As a proof of concept we apply this data compression technique to the current CMB data from WMAP , ACBAR , CBI , VSA and Boomerang . We find no evidence for the breaking of scale-invariance from measurements of four PCA mode amplitudes , which is translated to a constraint on the scalar spectral index n _ { { S } } ( k _ { 0 } = 0.04 Mpc ^ { -1 } ) = 0.94 \pm 0.04 in accordance with WMAP studies .