How to protect compiled Java classes?

First if you're targeting "only" the Windows market there's a very easy to prevent the ".class to .java" decompilation: use a tool like Excelsior Jet that will transform the .jar in an .exe.

This is foolproof: it is impossible to get the .java file back if you use Excelsior Jet (so long for all the people saying "it's impossible to prevent decompilation of a .class file"). Sure, an attacker could launch SoftIce and try to trace your .exe but that will prove a bit trickier than using JAD to decompile the .class to a .java and it certainly won't allow to find the .java file back.

Now maybe you're targetting OS X and Linux too or you don't have $$$ to shell off for Excelsior Jet.

I'm writing a commercial software written in Java. That software only makes sense if there's an Internet connection. Hence we "protect" our software, amongst other, by having part of the computation happening on the server side: we have several .class that won't work unless they're generated from the server side and we send them down the wire (and what is sent on the wire is always different: we're generating unique, one-off .class files on the server side).

This requires an Internet connection but if the user doesn't like how our software works then he's free to buy one our competitor's inferior product ;)

Decompiling will not do much good: you actively need to crack the software (ie reproduce what is happening on the server side) or you won't be able to use it.

We use our own "string obfuscation" before we use Proguard. We also do source code instrumentation (we could have done bytecode instrumenation as well) where we remove lots of things from the code (like the "assert" that we comment out) and introduce some random "code flow obfuscation" [the software can take different paths yet obtain the same result, this is something that really makes the software hard to trace]).

Then we use Proguard (which is free) to flatten all our OO hierarchy and to obfuscate the already-code-flow-and-string-obfuscated code.

So our flow is:

string obfuscation
random code flow obfuscation
Proguard
final .jar that depends on .class that are (differently) dynamically generated on the server side.

In addition to that we release very regular (and automated) update which always make sure to modify a bit our client/server protection scheme (so that with each release an hypotethical attacker has to start mostly from scratch).

Of course it's easier to throw the towel in and to think: "there's nothing I can do to make an attacker's life harder because JAD can find back the .java file anyway" (which is more than very debatable and blatantly wrong in the case where you use a .class to .exe converter to protect your .class from decompiling).

An obfuscator (see http://java-source.net/open-source/obfuscators) will "scramble" the code such that it will not make any sense when de-compiled.

There are several methods: