Alice has decided to stop using electronic communication with her friends and relatives. She believes analog media are generally more secure and private. Her friends and relatives, however, neither care about security nor want to be inconvenienced by it. Alice believes the postal service provides the most cost effective way to correspond. What steps can she take to prevent anyone but her intended recipients from reading her mail?
Alice is mistaken regarding analog media in this case. It's often overlooked, but because (by it's very nature) digital communication involves a computer at both ends, you can get the computer to do the work required for strong crypto, and so strong crypto becomes possible.
You can do secure communications over analog (typically with either one time pad or with a code) but these options require work to be done by the recipient.
For effortlessly secure communications you need strong crypto and that means computers.
A couple of ways that one might do this.
You could use Steganography. That is, you can try to hide your secret message some how. This is probably the most common way that prisoners behind bars attempt to communicate with the outside world or with other people in the prison. While it is hard (if not impossible) to prove the security of such methods, it is used every day with great success. If you, for example, don't have internet (or even a computer), then hiding your message would probably be a good idea.
You could combine the old with the new. You could write your message in a text document on a computer, then encrypt it, put the encrypted file on an SD-card and then mail the SD-card in the regular mail. You might even want to compute a hash of the whole card to verify that the card hasn't been tampered with. Now, if you are in prison, this might not be an option.
If you don't have a computer or the recipient doesn't have a computer, then you might consider a handcipher combined with hiding your message somehow. What is the strongest hand cipher you then ask? Good questions: https://crypto.stackexchange.com/questions/1653/what-is-the-most-secure-hand-cipher. Depending on how easily you can communicate with your recipient, you might even consider a one time pad (see also: https://crypto.stackexchange.com/a/8239)
Now, one might ask the question whether snal mail actually is more secure than electronic communication. Some will say that snail mail isn't more insecure than email, but all this depends on who your adversary is. If the Government want to read your regular mail, they can probably get a court order allowing them to do so. That said, this will involve more resources than just hacking your computer. If you are worried about the mass collection of data that the NSA is reportedly engaged in, then yes snail mail will be much more secure. The point is, that it all depends on who and what you are "afraid" of. It depends on your situation/ What country do you live in? Are you are spy communicating with a foreign government and is there a good chance that you are being monitored? If I, for example, wanted to hide communication from my neighbour, then I might not want to use snail mail because he might (might not) have easy access to my mail.
Send a piece of advertisement through snail mail. Then email them your encrypted message for which the key is some piece of the ad (eg. concatenate the fake serial no. of X items from page Y). I think the trick might be your contact inadvertently thrashing the advertisement! ;)
If we don't consider acceptable the requisite of OCRing a text -yet not dare send things unencrypted-, I think we should just use it as an external channel for key sharing. We could send CDs through the postal service, but somehow I think it would be more suspicious than papers.
Similarly, you could rely on security by obscurity and just hope that even if your was inspected, they didn't look into it so much that there was a hidden message written with orange-juice. That will obviously not pass any serious inspection.
On the other hand, if Alice only cares about transmitting a tiny bit of information (eg. "I arrived the safe house fine", "I am still in the run", "They caught Fred"), then a pre-established code where that data was concealed in normal-looking letters would be both suitable and preferable.
If both Alice and the intended recipient have mailboxes that lock, Alice can "trust the postal system".
Alice has a marginal guarantee that the mail man hasn't been paid off or isn't for some reason stealing or reading her mail. This is a laughable assumption but it's good enough for Alice's medical and banking records, legal documents, cable bills, utility bills, mail from her kids school, official election ballots. Quite frankly Alice and everyone else make this insane assumption every single day about data we really shouldn't make this assumption about. So before Alice goes any further Alice must have communication that is so far beyond all the stuff that regularly gets handled by the postal service.
Assuming Alice has really secret data. Alice can't trust the courier, she is faced with what cryptographers call the Byzantine Generals Problem. She can solve her problem using encryption. Here she has mad a baseless assumption that she cannot trust digital systems. There are analog encryption systems and even analog computer encryption systems. Alice's problem is that an algorithm that can be mathematically proven to be more secure than any other algorithm and provide security comparable to what is automatically done by gmail would be prohibitively tedious to do without a digital computer, with one exception. The exception is that Alice and her recipient use one time pads. This is tedious because she and her recipient must swap one time pads before they communicate by mail and one time pads cannot be reused. One time pads also have other disadvantages and vulnerabilities. Probably the biggest advantage is her recipient has to decode the message and keep track of a bunch of one time pads. Now if only there was a way for Alice to make an infinitely long one time pad that she could use for and infinite number of messages. Hooray, we have invented stream ciphers. Unless Alice and her intended recipient are math savants that's a lot of tedious math to do by hand.
Currently the US government uses AES. Which is what gmail uses. If she doesn't trust gmail she could trust some other program or write her own to encrypt the text before she puts it into gmail. But once again we're taking a simple problem and adding tedious levels of complexity.
Bottom line. Alice can "trust the postal system" if this concerns her she should use gmail and if this concerns her she should do all her communication in person in a locked vault in a secret location and if this isn't good enough for her she should see a doctor because she isn't communicating nuclear codes, she's crazy.