We examined the performance of several low-cost accelerometers for highly cost-driven applications in recording earthquake strong motion. We anticipate applications for such sensors in providing the lifeline and emergency-response communities with an immediate, comprehensive picture of the extent and characteristics of likely damage. We also foresee their use as 'filler' instruments sited between research-grade instruments to provide spatially detailed and near-field records of large earthquakes (on the order of 1000 stations at 600-m intervals in San Fernando Valley, population 1.2 million, for example). The latter applications would provide greatly improved attenuation relationships for building codes and design, the first examples of mainshock information (that is, potentially nonlinear regime) for microzonation, and a suite of records for structural engineers. We also foresee possible applications in monitoring structural inter-story drift during earthquakes, possibly leading to local and remote alarm functions as well as design criteria. This effort appears to be the first of its type at the USGS. It is spurred by rapid advances in sensor technology and the recognition of potential non-classical applications. In this report, we estimate sensor noise spectra, relative transfer functions and cross-axis sensitivity of six inexpensive sensors. We tested three micromachined ('silicon-chip') sensors in addition to classical force-balance and piezoelectric examples. This sample of devices is meant to be representative, not comprehensive. Sensor noise spectra were estimated by recording system output with the sensor mounted on a pneumatically supported 545-kg optical-bench isolation table. This isolation table appears to limit ground motion to below our system noise level. These noise estimates include noise introduced by signal-conditioning circuitry, the analog-to-digital converter (ADC), and noise induced in connecting wiring by ambient electromagnetic fields in our suburban laboratory. These latter sources are believed to dominate sensor noise in the quieter sensors we tested. Transfer functions were obtained relative to a research grade force-balance accelerometer (a Kinemetrics TM FBA-11) by shaking the sensors simultaneously on the same shake table and taking spectral ratios with the output of the FBA- 11. This reference sensor is said to have 120 db dynamic range (-+20 bits, though we only digitized it at 16 bits resolution and drove it with relatively small signals). We did not test temperature sensitivity, which is thought to be a significant issue at least for the silicon devices. Though these tests were not designed to be definitive (our anticipated applications do not demand research-grade precision), our tests do appear to have been successful in estimating relative transfer functions from about 0.3 to 50 Hz. Most sensors performed adequately in this range, with essentially fiat relative transfer functions. Noise tests appear to measure sensor noise well for the noisier (generally less expensive) instruments from about 0.1 to 50 Hz.