Modern analytical instrumentation is capable of measuring a variety of trace elements at concentrations down into the single or double digit parts-per-trillion (ng l-1) range. This holds for the three most common sample media currently used in environmental monitoring programs: filtered water, whole-water and separated suspended sediment. Unfortunately, current analytical capabilities have exceeded the current capacity to collect both uncontaminated and representative environmental samples. The success of any trace element monitoring program requires that this issue be both understood and addressed. The environmental monitoring of trace elements requires the collection of calendar- and event-based dissolved and suspended sediment samples. There are unique problems associated with the collection and chemical analyses of both types of sample media. Over the past 10 years, reported ambient dissolved trace element concentrations have declined. Generally, these decreases do not reflect better water quality, but rather improvements in the procedures used to collect, process, preserve and analyze these samples without contaminating them during these steps. Further, recent studies have shown that the currently accepted operational definition of dissolved constituents (material passing a 0.45 ??m membrane filter) is inadequat owing to sampling and processing artifacts. The existence of these artifacts raises questions about the generation of accurate, precise and comparable 'dissolved' trace element data. Suspended sediment and associated trace elements can display marked short- and long-term spatial and temporal variability. This implies that spatially representative samples only can be obtained by generating composites using depth- and width-integrated sampling techniques. Additionally, temporal variations have led to the view that the determination of annual trace element fluxes may require nearly constant (e.g., high-frequency) sampling and subsequent chemical analyses. Ultimately, sampling frequency for flux estimates becomes dependent on the time period of concern (daily, weekly, monthly, yearly) and the amount of acceptable error associated with these estimates.