A time series model was used to describe the daily milk yield of healthy co
tes in the first 48 days of lactation. A moving average (MA) model of order
I on the first-order differences of the data was selected - the same chose
n in a previous study by Deluyker et al. (1990). When the model was used to
generate predicted daily yields in a second data set, for cows in which cl
inical ketosis had been diagnosed, it was found that significant deviations
of actual yield below the daily forecast occurred from 3 days before the d
ay of diagnosis. The model appeared to be transportable to healthy cows fro
m another herd. Threshold values were defined to identify ailing animals by
their deviation from predicted yield. However, the thresholds were not ver
y sensitive, and required that a fairly high level of false positives be ac
cepted (25% of a healthy herd over a 5-day period).