Quantitative characterization of subsurface defects is a challenging problem in eddy current testing (ECT), not only due to the limitation of the penetration depth of eddy currents, but also due to the ill-posed nature of the inverse problem associated with quantifying a defect's 3D dimensions and its location in the test sample from measurements made on the sample surface. This paper presents a new ECT method for subsurface defect detection and quantification. Eddy currents with continuously varying frequency in a selected frequency band is induced in the conductive sample by using a chirp modulated excitation scheme. The frequency band of the chirp waveform is chosen so that the penetration depth of eddy current in the selected frequency band covers the thickness of the sample under test. The presence of a defect will disturb the distribution of the induced eddy currents. This disturbance is measured in terms of a varying magnetic field using an array of tunneling magneto-resistance (TMR) sensors. A C-scan image is obtained by linearly scanning the specimen. The footprint of the defect is calculated using spatial features of the C-scan image while the depth of the defect is estimated by analyzing the frequency characteristics of the measured chirp signal. A three-dimensional estimate of the dimensions of any subsurface defect can be obtained using the empirical procedure. The operating principle of this method is studied with a finite element method (FEM) model and then validated experimentally with a prototype array probe consisting of 180 TMR sensors. Thanks to its high sensitivity, the TMR sensors are suitable for measuring the weak low frequency magnetic field signal caused by a deep defect. The spacing between each two adjacent sensors is as small as 0.5 mm providing a magnetic field image with fine spatial resolution, which is beneficial for defect quantification. An aluminum sample with embedded defects was tested using the prototype probe. It is seen that the proposed scheme utilizing spatial and frequency domain features is capable of quantitatively characterizing subsurface defects.