Abstract Diagnostic sampling, or surveying, of operating concentrators – with a view towards identifying flowsheet improvement opportunity – has long been a valuable field of endeavour, but has seen little publication. A quantitative protocol entitled Statistical Benchmark Surveying has been developed at Falconbridge, and was tested at the Raglan mine in Northern Quebec. Several prototypes were tested, leading to this current proposal. It is shown that the field of applied statistics brings value to this approach by using the appropriate distribution models. Replicate two-hour survey units are taken. This engages the powerful averaging effects of the Central Limit Theorem. The effects of auto-correlation in time are broken by spacing the surveys with knowledge of the semivariogram. This importantly moves the sampling from auto-correlated to random. Use is made of the paymetal grade measurements (in this case nickel) to construct a system of reference distributions. This recognizes the dominant influence of ore grade in concentrator performance. Outlier rejection from the set of six replicate two-hour surveys is proposed by these reference distributions, which are designed to operate at the 95% confidence level. A composite is prepared from those surveys which pass these tests. As a result, this flotation feed composite has a mean paymetal grade close to that typically experienced at the operation. The flotation feed and associated suite of flowsheet composites are used for a mass and value balance, and are later presented to Qem∗SCAN for quantitative mineral measurement as a series of sized fractions.