Instead you can create yourself your privkey (at least the hex one, 64 characters long, I don't know if all base59 ones are valid they are not because of the checksum) using your own pattern that you know by heart, thus no need to write it
E.g. 1234567890abcdef1234567890abcdef1234567890abcdef1234567890abcdef for 19ffB4HttNCHfY1t3YuErEytCspyHyVMwv

#> bitcoin importprivkey 1234567890abcdef1234567890abcdef1234567890abcdef1234567890abcdef
error: {"code":-5,"message":"Invalid private key"}

Afaik, importprivkey only accepts base58 privkeys


You can use any 256bit number as private key, except 0

You can use any 256bit number as private key, except 0

To those saying a pass-phrase based key can't be used because it lacks entropy have obviously never heard of Password Based Key Derivative Functions

http://en.wikipedia.org/wiki/PBKDF2

To those that haven't heard of key derivitives they work because instead of taking a hash of the passphrase (i.e. "This is my secure bitcoin passphrase"),  you take a hash of the hash of the hash of the hash of the hash of hash ..... (n iterations later) of the passphrase.

Also a random salt is chosen at the time of password generation and is added to each round of hashing. 

To make it brute force proof one simply needs to pick a large enough n so that it takes a "non trivial" amount of time to hash one password.

Current top of line GPU can perform < 1 GH/s. 
So you make n = something in the magnitude of 10 million.


An example program would prompt the user for 4 randomly chosen words (w1,w2,w3,w4) and a 4 digit number (p).  (You could have the program not generate a key for words which are too common to improve security).
passphrase = w1+w2+w3+w4
p = salt
n = p * 2^12  (the 2^12 is a constant to match all possible 4 digit p's to a n in the range we need (~10M so it takes a GPU a "non-trivial" amount of time to complete).
 
Now simply run a key derivative function which uses a pass-phrase of "w1+w2+w3+w4" iterates n times with each round salted with p.
...
a single 1 GHash/s GPU making 30 attempts per second would take nearly 11 million years to try all possible combinations of passwords and salts.
However that is just one GPU.  What if deepbit pool tried to brute force the key?  With 5THashes/s of computing power it would only take deepbit ~3,400 years.  Another way to look at it is the entire deepbit pool would need to work for 34 years 24/7/365 just to have a 1% chance of breaking the private key.

To regenerate the private key in the future the user would simply need to remember the 4 words and 4 digit pin.  Hopefully this gets some people thinking.