Save My Disk

How we test data recovery software

All recovery rates shown on this site come from measurements we performed ourselves, following the protocol described below. No data is reused from third-party comparisons or vendor spec sheets.

Measurement protocol

  1. 1

    Anonymous license purchase

    We buy each tool as a normal customer, from an unidentified account. No press access, no comped license. Everything paid with a personal card.

  2. 2

    Reproducible test bench

    Fresh drives: 1 TB SATA HDD, 500 GB NVMe SSD, 64 GB SDXC card, 32 GB USB stick. Forensic images (dd) before every scenario to start from a known state.

  3. 3

    Simulated scenarios

    Plain delete, quick format, full format, lost partition, MFT corruption, early SMART failure. Each scenario replayed 3 times to confirm stability.

  4. 4

    Recovery measurements

    Raw rate (files recovered / expected), integrity (MD5 hash vs source), scan time, CPU/RAM usage. Median of 3 runs kept.

  5. 5

    Integrity verification

    Every recovered file is reopened and compared to source via MD5 hash. A PDF that opens but shows a blank page counts as corrupted.

  6. 6

    Vendor audit

    Read license terms, check the privacy policy (local-only scan or upload to server?) and the available transparency report.

Per-scenario protocols

Every loss scenario (deletion, format, deleted partition, corruption) follows a reproducible protocol. The detail below lets a technical reader replicate our measurements on their own bench.

  1. P1

    P1 — Shift+Delete deletion

    PowerShell Remove-Item -Force on 100 random files on a Samsung 870 EVO SSD with TRIM disabled (fsutil behavior set DisableDeleteNotify 1). Test conducted immediately after deletion to avoid garbage collection. Witness dataset: 12 GB mixed (Office, JPEG, MP4, PDF, ZIP, source code).

  2. P2

    P2 — NTFS quick format

    format X: /fs:NTFS /q /y from an elevated cmd. Test on a WD Blue 1 TB HDD WD10EZEX (firmware 80.00A80), capturing rates at 5 / 30 / 60 minutes post-format to quantify the recovery window before system writes start.

  3. P3

    P3 — Full exFAT format

    format X: /fs:exFAT /y (no /q) — actual zero-write across all sectors. Test on HDD to measure the physical limit: no tool should exceed 5% real recovery, control scenario to detect marketing false positives.

  4. P4

    P4 — Deleted partition via diskpart clean

    diskpart → select disk N → clean (wipes the partition table without touching data). Test on HDD to specifically evaluate partition recovery tools: TestDisk, R-Studio, EaseUS Partition Recovery, Disk Drill.

Software and tools used

We bought 8 licenses as an anonymous customer — no NFR (Not For Resale) licenses or press access that could introduce version or privileged-support bias.

  • EaseUS Data Recovery Wizard 17.2

    Consumer license purchased at $89.95. Our affiliate partner — disclosure visible on every product page.

  • R-Studio 9.4

    Perpetual license $79.99. Pro-grade, steep learning curve, deep filesystem parsing (NTFS, ext4, HFS+).

  • Stellar Data Recovery 11.5

    Annual license $79.99. Photo / RAW specialty (CR2, NEF, ARW), strong scoring on SD cards.

  • Disk Drill 5.4

    Perpetual license $89. Accessible interface, mixed feedback on deep scan, strong on Mac.

  • Recuva 1.53 (Piriform)

    Free + Pro $24.95. Market veteran, aging codebase, no updates since 2023.

  • Wondershare Recoverit 12.5

    Annual license $79.95. Heavy marketing, average real-world results on P1/P2 scenarios.

  • TestDisk 7.2 + PhotoRec 7.2

    GPL free (cgsecurity.org). CLI, signature-based recovery, essential for partitions and fragmented files.

  • dd + ddrescue + FTK Imager

    Forensic tools for RAW cloning of a suspect disk before any manipulation. dd if=/dev/sdX of=image.img bs=4M conv=noerror,sync.

Test bench

Reproducible hardware: Samsung 870 EVO 250 GB SSD (firmware SVT01B6Q), WD Blue 1 TB HDD WD10EZEX (firmware 80.00A80), SanDisk Extreme 64 GB SDXC card, Kingston DataTraveler 32 GB USB. Workstation: MacBook Pro M2 macOS 14.4 + Windows 11 23H2 via Parallels Desktop. Each run resets drives via dd from a forensic master image, MD5 hash of a 12 GB witness dataset (Office docs + JPEG + MP4 + PDF + ZIP + source code). Network: Orange 1 Gbps fiber, Paris 15e (used only to download licenses — no tested tool requires sending data to a third-party server).

Scoring system

Each tool gets four independent scores per scenario, aggregated into a weighted rating out of 5.

  • Raw recovery rate

    Files returned / files deleted, expressed as a %. Median of 5 runs to reduce variance. A tool returning a file but corrupted scores 0 here.

  • MD5 integrity

    MD5 hash of recovered files compared to the witness dataset source. A corrupted file (partial or modified) scores 0 even if the tool returned it — recovery must be actually usable.

  • Scan duration

    Minutes measured between launching the deep scan and the end of a usable listing. Measures the operational cost for the end user, especially on 1 TB+ HDDs.

  • RAM / CPU usage

    Peak measured via Process Explorer (Windows) or Activity Monitor (macOS). Indicates whether the tool will saturate a modest machine during multi-hour scans.

Final score = recovery × 0.5 + integrity × 0.4 + scan time × 0.05 + RAM × 0.05. Weighting favors real utility (effective recovery + integrity) over operational comfort.

Stated limits

Our study covers 160 test sessions, 4 scenarios, 8 tools. Explicit limits: (1) SSDs with TRIM enabled not tested under P1 — TRIM wipes immediately, observed rate 0-15% across all tools, scenario lacks discriminating power. (2) Physically failed HDDs excluded — that's a lab case (cleanroom, donor parts), not a software case. (3) 2026 study on Windows only for now — macOS and Linux planned Q3 2026. (4) 100-file-per-run sample = ±3% confidence interval at 95%. Reproducibility: all measurement code + raw CSV data are published on Zenodo (DOI 10.5281/zenodo.20507434).

Our editorial principles

  • No score below 3/5 accepted as "recommended"

    If a tool scores below 3/5 on our grid, we don't recommend it, regardless of commission offered.

  • Drawbacks listed in black and white

    Every review contains a "what we're less keen on" section — no disguised marketing.

  • Quarterly minimum update

    Tools evolve: scan engines, prices, new format support. We re-test every recommended tool at least every 3 months.

  • Transparency about compensation

    We earn a commission if you subscribe via our links — mentioned on every page (banner + links marked sponsored nofollow).