RAID levels

In RAID 5, if any two disks fail, your data is LOST COMPLETELY.

In RAID 6, you can have two failures and your data will be fine.

In RAID 10, You MIGHT have your data, depending on which two drives fail.

since the question seems aboot keeping data safe, don't forget that even with RAID systems, your best protection is a backup
if you need the whole system - use an image backup !

Raid 0 no tolerance for any failure, data lost.
Raid1,5 tolerance for a single failure, a second failure makes data inaccessible/lost
Raid 6,10,50(51) 6 can tolerate any two drive failures, raid 10 can tolerate a single drive failure in each raid 1 group, raid 50 (mirror of raid 5) can tolerate a single disk in each of the raid 5 groups.

There is raid.org I think that covers in much greater all aspects of raid including performance/costs.

One vital bit of information the others missed.  I'll get on my soapbox a bit because I haven't reminded the good experts here of things they already know, but are dangerous not to mention because people get a false sense of security..

The responses fail to let you know that if you are degraded then you can still have partial data loss if there is either an XOR parity error, OR any of the surviving disks have unreadable blocks.

Unfortunately this is quite common and why one needs to do consistency checks and backups.

To be clear, lets say you have a 5 drive RAID5 using SCSI drives that has worked flawlessly last 10 years, and you never ran a consistency check because your crappy controller doesn't do that automatically, or you just never knew.

You lose a disk, so your hot spare kicks in and it starts to rebuild extrapolating data from other drives via XOR so the new disk gets stuffed.  For every stripe of data (horizontal slice) prior to the failure you would have had one drive that actually mapped to the block 0 your O/S read, and in that same stripe you would have had one drive that contained parity (the redundant XOR).

Up till now the parity drive for that slice hasn't been read or written to for last 10 years because the drive that really had the data was working fine, and, lets face it, it is rare that you update the partitioning.  

Since you now have degraded slice block 0 has to be extrapolated.  You had a bad block.  The controller will throw an error, and take the 2nd disk off line ...

Result ... at worst, 100% data loss of everything, unless you pay $5000 bare minimum (going rate isn't going to be less than $1000/drive for RAID recovery, but I've seen it go higher than $3000 per drive).

At best, you have a high-end controller and it leaves you with a bunch of 0s for the first slice.   But that means all your directory information is gone, so not only do you lose all partitioning, but you'll lose the file names or at least a high percentage of file names, but at least you can buy software that can recover.

Bottom line, real world, RAID5 can't be trusted to prevent data loss if you lose one disk, and if you aren't doing consistency checks, you need to.  The consistency checks reads all blocks on all disks and would  have REPAIRED that unreadable block mitigating risk.

Dont do RAID5, risk of data loss is actually higher than RAID0, because with RAID5 a partial rebuild on live system that gets a read error is going to screw things up so badly that realistically you are looking at data recovery costs near the price of buying a car.  At least with RAID0 and you lose a block or even drive, it damage is going to be limited to just whatever failed.

Go RAID6 and RAID1/RAID10, RAID5 was great when disks were limited to a few hundred GB because rebuilds wouldn't take more than a few hours..  Now statistically the error correcting rates in drives make it a statistical certainty that you'll have at least one just from the amount of data that you have to read/write to rebuild an array using consumer-class multi-TB disks.