I have used this instrument and my graduate students. SPAD doesn't measure chlorophyll content of the leaf,but it's reflection with light!, e.g, the correlation between spad readings and chlorophyll content is not linear !! .So, they the readings are just absolute numbers without units! . You need 4 to 5 readings on different leaves, upper, middle, lower,..and get the mean of each ,and compare. Luck.

Always aim for n = 30 at all levels of replication to fulfil the CLT :)

It depends on what you mean by "good result" doesn't it?

If you just need a value to do some quick calculations, Medhat Elsahookie 's 4 or 5 samples will (probably) suffice. On the other hand, if you really need to know the variability inherent in the leaves from up to 10 plants per pot, perhaps Andrew Paul McKenzie Pegman 's n = 30 is closer to the mark (maybe it will take even more than that!).

Why not plot the standard deviation of the average vs the number of samples and see if there is an asymptote. At some point more data (greater n) will not improve the estimate of variability.

I would probably also suggest investigating a more direct measure of chlorophyll by extracting leaves in methanol for spectrophotometry. There are numerous methods searchable on ResearchGate and elsewhere on the web.

It depend on the purpose of your research, if you are just comparing millet genotypes in response to an environmental effect, doing 3 to 5 measurements on different plants representative of the plot is enough (suppose also, you have a good homogeneity of plants inside a plot) . The measurement (1) can be done in the middle of the latest fully expanded leaf (generally, the 3rd leaf from the top). Avoid some statement like "measure chlorophyll content using spad", it doesn't actually measure the chlorophyll content, but you can say "relative leaf chlorophyll content".