This talk is to initiate research and discussion of the problem of climate normals for a changing climate. Analyses of climatic records show that contemporary climate variations contain non-stationary components (climatic trends) which need to be accounted for in evaluating and predicting climatic normals and climatic anomalies. The first attempt to consider the existence of climatic trends was the WMO recommendation that climatic normals be recomputed each decade by averaging observations for the last three decades. Because a trend related change in a climatic variable during one or a few decades often exceed the standard deviation of the variable the WMO normals cannot be used in statistical climate predictions. Consequently, scientists at the Climate Prediction Center (NCEP/NOAA) exploited the concept of “Optimal Climate Normals (OCN)” which are 10-year averages for temperature and 15-year averages for precipitation. However, OCN is a compromise that does not solve the problem of very strong climatic trends and has large sampling errors because of its short averaging times. An alternative solution to this problem is to use empirically estimated climate trends and time dependent climatic normals. A new technique has been recently developed and tested for approximating seasonal and diurnal cycles in climatic trends. The technique allows one more efficiently and accurately than previous methods to separate the problem of extrapolating the expected value (climatic trend with its seasonal and diurnal cycles) from the problem of extrapolating the random and/or quasi-periodic (like ENSO) components of climatic processes (detrended climatic anomalies or residuals). Here we use a simple statistical model to analyze errors of different approaches for evaluating and extrapolating climatic normals into the future. It looks as if the WMO technique was a good treatment for very weak climatic trends. The OCN technique is suitable for medium trends but is not adequate for strong climatic trends. We suggest using the observed data to estimate climatic trends and to utilize estimated dependence of expected value (climatic normals) on time. Such an approach should work properly even for very strong trends with a reasonable trend model.