International agreements such as the EU Water Framework Directive (WFD) ask for efficient sampling methods for monitoring natural resources. In this paper a general methodology for designing efficient, statistically sound monitoring schemes is described. An important decision is the choice between a design-based and a model-based method, implying the choice between probability (random) sampling and purposive sampling. For mapping purposes, model-based methods are more appropriate, whereas to obtain valid results for the universe as a whole, such as in testing water quality standards against legal standards, we generally prefer a design-based method. Four basic sampling patterns in space-time universe are described: static, synchronous, static-synchronous, and rotational. A case study is carried out for monitoring the quality of surface water at two farms in western Netherlands, wherein a synchronous sampling design is applied, with stratified simple random sampling in both space and time. To reduce laboratory costs the aliquots taken at the locations of a given sampling round are bulked to form a composite. To test the spatiotemporal mean N-total concentration during the summer half-year against the MAR standard with a power of 80% at a concentration 15% below the MAR standard and with a confidence of 95%, six to nine sampling rounds are needed with 50 to 75 locations per sampling round. For P-total the required number of sampling rounds differs strongly between the two farms, but is for both farms much larger than for N-total.