Yun-Ru is right. For me it depends also on the equipment you have, the efforts you'd like to put in but also on the strength signal that you expect, is the gene highly expressed? Additionally, RNA duplexes can be washed more stringently, increasing specificity and signal to noise ratio which is allways the bottleneck of a technique.
Easiest way for me is still random-primed p32 DNA probes.
Benjamin is right but you can also do single stranded DNA probes using a PCR-Product as template and adding just one Primer to a "PCR" reaction (linear increase of your Product). Radioactive Probes are very sensitive, but I had some very good results with DIG-dUTP labeled Probes. The advantage of DNA especially dsDNA-Probes is that you can use them 5-10 Times, which is not recomendable with RNA-probes.
I have done hundreds of Northern blots and generally use dsDNA probes. The main reason is that the labeling is easier. I use random primed labeling with 32-P-dCTP. Also, for RNA probes, background can be very high and people often use RNAse in the washing to reduce the background. This makes it impossible to re-probe the same blots later.