In this paper, we rigorously compare compressive sampling (CS) to four state of the art, on-mote, lossy compression algorithms [K-run-length encoding (KRLE), lightweight temporal compression (LTC), wavelet quantization thresholding and run-length encoding (WQTR), and a low-pass filtered fast Fourier transform (FFT)]. Specifically, we first simulate lossy compression on two real-world seismic data sets, and we then evaluate algorithm performance using implementations on real hardware. In terms of compression ratios, recovered signal error, power consumption, on-mote execution runtime, and classification accuracy of a seismic event detection task (on decompressed signals), results show that CS performs comparable to (and in many cases better than) the other algorithms evaluated. A main benefit to users is that CS, a lightweight and nonadaptive compression technique, can guarantee a desired level of compression performance (and thus, radio usage and power consumption) without subjugating recovered signal quality. Our contribution is a novel and rigorous comparison of five state of the art, on-mote, lossy compression algorithms in simulation on real-world data sets and in implementations on hardware.