The accurate prediction of lactation yield for a dairy cow has been a long standing goal for improving management. Historically the approach has been to assume that yield as a function of time follows a lactation curve which can be assigned a form that applies to all cows, and that cows differ in the parameters of this common curve. This has led to a plethora of forms being suggested for this common curve, of which perhaps the most well-known is that of Wood(1967). The biggest problem with this approach is the assumption that all cows and all management systems combine to produce individual lactations described by a single form of ‘lactation curve’. A novel approach using mixed-model smoothing was investigated in this study which: (i) fits a shape of lactation, assuming no specific parametric form , to daily yield records of a group of cows by using a patterned (co)variance matrix to describe deviations from a linear trend for time postpartum in a mixed model; and (ii) allows individual yields to be predicted using this ‘mean’ curve and the deviation of an individual's yield record from this mean.