We present a novel multimedia data hiding system using over-complete transform dictionaries. The proposed design scheme is different from most existing schemes that use orthonormal (complete) dictionaries, e.g., discrete Fourier transform (DFT), discrete cosine transform (DCT), or discrete wavelet transform (DWT) matrices, etc., during data embedding and detection phases. This paper focuses on one of the advantages brought by the use of over-complete dictionaries, i.e., substantially enhanced system security as compared to conventional systems. Specifically, a spread spectrum (SS)-like system is designed and compared to its classical form. The over-complete dictionary is partitioned into two blocks and applied to host signal and embedded data respectively. By appropriately designing the dictionary and partition mechanism, we illustrate that without compromising the performance on embedding distortion, benchmark robustness, and embedding capacity, the proposed system can achieve a high level of security by randomizing several designing parameters. As a result, the system is still computationally efficient, but unauthorized detection of embedded data could become impractically difficult. Experimental examples using image data are provided to evidence our proposal. The proposed system is also applicable to other multimedia formats.