Yield responses to the application of foliar fungicides to wheat are highly variable. As the cost of treatment is known, some method of estimating future yield response is required if treatments are to be applied only when the value of the yield beneรt will exceed the cost. Treatment decisions are often supported by disease thresholds, either formalised or developed by crop managers through practical experience. In farm practice, the proportion of yield that is due to fungicide treatment is not usually known, so the success of a spray decision is often judged by the level of disease later in the season. This paper presents data from รeld experiments in 1994 and 1995, showing that variation in the current level of visible disease (yellow rust and Septoria tritici) explains little of the variation in future e โ ects of the disease on the host, and that the yield e โ ect of a unit disease is not constant across sites and seasons. These รndings suggest that traditional disease thresholds may be unreliable predictors of the need to spray and that estimating the success of a treatment decision by observing disease levels later in the season is prone to error. A "negative thresholdร, deรned as that level of disease below which an economically damaging epidemic cannot develop within a known time-period, may be more reliable. Below the negative threshold, treatment is not required. Above the negative threshold, other factors a โ ecting the rate of epidemic development and sensitivity of the host to green leaf area loss need to be considered, in order to quantify the need for treatment. Measurements which reรect the cropรs ability to intercept solar radiation may prove more reliable tools than percentage disease for judging the success of treatment decisions and, experimentally, for quantifying the e โ ect of variation in risk determinants.