A mirror gives you two copies of everything — which is exactly why RAID 1 recovery is usually the most winnable of all the RAID jobs, and exactly why the wrong rebuild can throw that advantage away.
Where one member can’t read a sector, the other usually can. The job is to merge both into a single complete image and rebuild the file system from that — not to rebuild the array onto a disk that may be failing too.
RAID 1 writes the same data to two or more disks at once. Lose one and the array carries on from the other, which is why it is the simplest and most robust of the common levels. What it protects against is a disk failing. What it does not protect against is anything that writes bad data to both members at the same moment — a deletion, a ransomware run, a corrupt file, a controller writing garbage. Both copies dutifully receive it. A mirror is redundancy, not backup, and the distinction matters most on the day you need it.
The common surprise with RAID 1 is that both disks tend to degrade together. They are usually the same model from the same batch, installed on the same day, and they have handled an identical write workload ever since — so when one develops bad sectors, the other is frequently not far behind. That is why a mirror can appear healthy for years and then present two failing members within weeks of each other. The other failure mode is silent: one disk drops out, nobody notices because the volume still mounts, and the array runs unprotected until the survivor fails too.
Each member is imaged read-only on hardware built for failing media, with retries limited and unreadable sectors logged rather than allowed to stall the clone. Then the images are merged: for every sector, whichever copy read cleanly is used. Because the two disks rarely fail in exactly the same places, the composite is usually complete even though neither individual disk is. The file system is rebuilt from that merged image, and the original disks are never written to at any point.
Where a mirror was managed by a controller or by Windows dynamic disks, the metadata offset is accounted for so the volume is read at the right boundary — a detail that silently defeats a lot of DIY attempts.
First, don’t force a rebuild when either member is suspect — that is the single most common way a recoverable mirror becomes an unrecoverable one. Second, don’t re-insert a disk the controller has already ejected; if it re-syncs in the wrong direction it will overwrite the good copy with the bad one. Third, don’t run repair utilities such as chkdsk against the array while it is degraded, because a write to a mirror is a write to everything you have left. If the data matters more than the uptime, power down and image.
Image each member read-only, then merge the images so that wherever one disk failed to read a sector the other supplies it. The file system is rebuilt from that merged image rather than from either disk directly. Doing it this way means neither original is written to, so nothing you try closes off a later option.
Usually, if the surviving member is genuinely healthy — a rebuild onto a fresh disk is a normal operation. The risk is rebuilding when both members are degraded: the process reads the survivor from end to end, and a failure partway through can leave you with two damaged disks instead of one. If either disk shows bad sectors or SMART warnings, image both before rebuilding.
It is intact, but it is no longer protected. A degraded mirror has no redundancy left, so any fault on the remaining disk means total loss. Copy the important data off to separate media before you replace anything, and treat the array as a single point of failure until the rebuild has completed successfully.
RAID, NAS and server recovery starts from £500 plus VAT, confirmed in a written quote after a free 48-hour diagnostic. Where a member needs physical drive-level work we take a 50% deposit up front, with the balance payable only on a successful recovery.
Both disks, labelled with which bay each came from, and a note of the controller or whether it was a software mirror. Don’t rebuild, don’t re-insert an ejected disk, and don’t initialise anything. If the unit is an enclosure, sending it whole is fine.