Many wireless sensor networks require the acquisition and transmission of high-precision analog sensor outputs. To force analog data into digital transmission systems inevitably causes irrecoverable quantization errors, (significant) bandwidth expansion, and the “significant-vs-insignificant bit” problem. This paper proposes a new transmission paradigm that exploits a single analog error correction code in lieu of quantization, digital error correction code and digital modulation. The novelty lies in the effective construction of nonlinear analog codes using appropriate chaotic functions through a parallel concatenation structure. Specifically, by carefully designing a triple-branch tail-biting baker's map code, and developing its maximum-likelihood (ML) decoder, we show that analog error correction can be achieved with high-performance and low-complexity. Simulations show that the proposed analog code can actually outperform the state-of-the-art digital scheme involving digital turbo codes and unequal error protection, on both AWGN and fading channels!