sharding versus replication for database scalability?

So, I'm trying to figure out the best way to scale our main user database, and I'd love to hear what you all think about sharding versus replication. Read traffic is definitely increasing, but writes are starting to pick up too. Replication feels like a no-brainer for handling all those reads, but how do you guys deal with the headache of sharding when writes become a problem? When does one option just make more sense than the other? I'm really interested in what you've seen out there and any mistakes to avoid.