How accurate is MyFitnessPal, according to published studies

Short answer

Published MyFitnessPal studies disagree on calories and nutrients, with closer energy results in controlled or cleaned records. Evenepoel 2020 after removing implausible entries, Lozano 2024 with a weighed meal and Diktas 2025 with provided foods found calories close to their references. Chen 2019, Banal 2024 and Morello 2025 found calories or nutrients off in independent logging or national food table comparisons, while Griffiths 2018 found protein, fat, sodium and cholesterol lower than the reference. Sodium and cholesterol were weaker in Griffiths and Evenepoel, and fiber was weaker in Lin 2022, Evenepoel and Morello.

How the food database is built

MyFitnessPal explains its check mark as a sign that it reviewed or added a food and believes the nutrition is accurate and complete. It describes using "a variety of tools and resources" but gives no further review method. An entry without the mark may be correct, and an entry with it may contain errors.

Both MyFitnessPal and its members add foods. A member can share a new food from the website to the public database. Creating a food for an unmatched barcode links that code to the food for all members.

To flag a problem, use the food entry reporting flow and identify the issue, such as wrong nutrition, a duplicate or a missing serving size. The nutrition team reviews the report and changes the entry if it confirms the error. Corrections are not instant.

MyFitnessPal publishes different database counts. Its check-mark help page says 19 million, while its homepage says both 20.5M and over 18 million.

The studies

These summaries come from each study's published abstract, without a full-text reading.

Griffiths et al., 2018

Public Health Nutrition. DOI for Griffiths et al., 2018.

A researcher entered 30 dietary recalls, each covering 24 hours from one of 30 lower-income adults, into five free apps. The reference was NDSR research software. This tested researcher-entered records, not people selecting their own entries.

MyFitnessPal reported significantly less total protein, with P of 0.03, total fat, with P of 0.005, sodium, with P of 0.02, and cholesterol, with P of 0.005. Across all apps, correlations were 0.73 to 0.96 for energy and macronutrients and 0.57 to 0.93 for sodium, total sugars, fiber, cholesterol and saturated fat. Those ranges are not MyFitnessPal-only results.

Teixeira et al., 2018

Nutrition and Dietetics. DOI for Teixeira et al., 2018.

In Brazil, 30 university students kept paper food records and entered intake in MyFitnessPal. Researchers compared energy, macronutrients and fiber with the paper records calculated using Brazilian food composition tables.

The abstract reports moderate positive correlations across variables and non-significant associations for energy and fiber. It describes a tendency to underestimate nutrients, with relatively narrow limits of agreement, but increasing overestimation of carbohydrates and lipids at higher intakes. The authors judged relative validity good, especially for energy and fiber, while warning about database gaps. The abstract supplies no numerical correlation values. Its reference and student sample were specific to this Brazilian study.

Chen et al., 2019

Nutrition. DOI for Chen et al., 2019.

In Australia, 43 adults who were not regular MyFitnessPal users logged on their own for 4 consecutive days. Researchers compared those logs with two researcher-run recalls covering 24 hours each.

People omitted a mean of 18% of food items, with a standard deviation of 15. Mean energy was underestimated by 1863 kJ, with a standard deviation of 2952 kJ and P of 0.0002. All macronutrients were underestimated too. The reported problems included matching foods, estimating portions and time. These results include omissions and portion choices, so they do not measure database quality alone.

Evenepoel et al., 2020

Journal of Medical Internet Research. DOI for Evenepoel et al., 2020.

In Belgium, 50 participants completed a 4-day MyFitnessPal record twice, 1 month apart. They received careful instructions. The reference was manual calculation with the Belgian Nubel database. The first record set informed limits for removing implausibly high entries from the second, rejecting 2.8% of its entries.

After that removal, correlations were 0.96 for energy, 0.90 each for carbohydrates, fat and protein, 0.80 for fiber and 0.79 for sugar. Cholesterol and sodium correlations were weak at 0.51 and 0.53. There was fixed bias for fiber and proportional bias for cholesterol, with no bias otherwise. The favorable energy result applies to cleaned records, not every entry participants originally logged.

Lin et al., 2022

Journal of the Academy of Nutrition and Dietetics. DOI for Lin et al., 2022.

Researchers compared four apps with NDSR version 2017 for 50 frequently consumed foods drawn from a 2017 study. The paper focused on mostly unprocessed and minimally processed foods, rather than a complete everyday logging routine.

The abstract reports good to excellent agreement for MyFitnessPal and Lose It!, with an intraclass correlation range of 0.89 to 1.00, except MyFitnessPal fiber at 0.67. Yet its Bland-Altman analysis found MyFitnessPal had the poorest energy agreement among the tested apps, with a mean difference of 8.35 kcal and standard deviation of 133.31 kcal. The shared correlation range is not an accuracy percentage for MyFitnessPal.

Banal et al., 2024

BMJ Nutrition, Prevention and Health. DOI for Banal et al., 2024.

The study population was 37 Filipino adults with obesity. Participants logged 5 days in MyFitnessPal and on paper. Three dietitians re-entered the paper records in MyFitnessPal, while a researcher calculated intake with Philippine Food Composition Tables.

Energy, protein and fat differed significantly from the tables, with very weak to moderate agreement. MyFitnessPal underestimated energy, carbohydrates and fat and overestimated protein. Dietitians had good agreement with one another except for fat. Agreement between people entering records did not establish agreement with the reference. The abstract gives no MyFitnessPal error percentages, and the result concerns this population and Philippine tables.

Lozano et al., 2024

American Journal of Clinical Nutrition. DOI for Lozano et al., 2024.

A randomized crossover lab study gave 43 participants a preweighed plated meal and leftovers. App estimates were compared with weighed food using equivalence bounds of 21% in either direction.

MyFitnessPal overestimated energy by 7.0% and met those equivalence bounds, with P of 0.04. The abstract result cited here gives no MyFitnessPal nutrient-by-nutrient figures. It covers a single lab meal, not unrestricted daily logging.

Diktas et al., 2025

Journal of Nutrition. DOI for Diktas et al., 2025.

This semicontrolled crossover study analyzed 44 adults. Each day, participants received a cooler of preweighed foods and ate only those foods. After training, they used each app for 3-day periods. The reference was weighed intake, with equivalence bounds of 18% in either direction.

MyFitnessPal energy estimates were equivalent to weighed intake, with P below 0.001 and mean percent error of 3.7%. The abstract result cited here gives no MyFitnessPal nutrient-by-nutrient figures. Training and the restriction to provided foods limit what this says about independent food choices and logging.

Morello et al., 2025

Journal of Human Nutrition and Dietetics. DOI for Morello et al., 2025.

Two raters entered 43 three-day food records from Canadian endurance athletes, 27 men and 16 women, into MyFitnessPal and Cronometer. The reference was ESHA Food Processor using the 2015 Canadian Nutrient File.

MyFitnessPal had poor validity for total energy, carbohydrates, protein, cholesterol, sugar and fiber. Its raters were inconsistent for sodium and sugar. The authors suggested the many entry choices, including unverified member entries, as a reason. The abstract gives no MyFitnessPal error percentages. This was a rater-entered comparison against Canadian reference data.

Cronometer had good to excellent agreement between raters for all nutrients and good validity except for fiber and vitamins A and D in this study. That is a separate result for Cronometer, not evidence about MyFitnessPal.

Where they agree and where they differ

The reference and logging conditions differ. Lozano used a lab meal, Diktas trained people and supplied their foods and Evenepoel removed implausible entries. Those studies found closer energy agreement. Chen observed independent logging, while Banal used Philippine tables and Morello used Canadian reference data, and found energy or nutrient problems. This comparison describes the study designs, not a conclusion attributed to any single author.

Teixeira reported good relative validity for energy and fiber, but Lin found weaker fiber agreement and Morello found poor fiber validity. Evenepoel found fiber bias and weak sodium and cholesterol correlations, while Griffiths found sodium and cholesterol significantly lower. The findings do not reduce to one accuracy percentage.

Tosi 2021 and Bzikowska-Jura 2021 also included MyFitnessPal among several apps, but their abstracts give no MyFitnessPal-only figures. Tosi DOI. Bzikowska-Jura DOI.

Barcode Scan, Meal Scan and Voice Log

Barcode Scan, Meal Scan and Voice Log all require Premium. The barcode help article makes no accuracy statement. The homepage calls Barcode Scan "fast, accurate", without supplying a test or number.

For photos, the Meal Scan help page calls the result "a starting point, not a final answer" and supplies no accuracy figure. Voice Log searches the database used for typed searches and makes no accuracy statement. See Cal AI and MyFitnessPal photo logging for more on those choices.

None of the studies above tested Barcode Scan, Meal Scan or Voice Log. The MyFitnessPal Premium cost guide covers what the subscription costs.

How to check one entry yourself

1. Start with the package label and compare it with the entry, including the portion.
2. For food without a label, search USDA FoodData Central, the USDA food composition resource with several distinct data types.
3. Choose a check-marked entry when available, while remembering that MyFitnessPal says it can still have mistakes.
4. Use a food scale to weigh your portion. Chen 2019 identified portion estimating as a problem participants had.
5. Send an incorrect entry through MyFitnessPal's report flow so its nutrition team can review the problem.

Trackers that show where each number came from

Cronometer names NCCDB and USDA SR28 among its data sources. Its curation team reviews member submissions before adding them. Company pages checked October 5, 2026.

MacroFactor's help center describes around 1,360,000 verified searchable foods in its standard database and 1,040,000 in its legacy database. Around 26,500 standard database entries are common foods with complete micronutrient data, with much of that common-food count coming from NCC. Help center checked October 5, 2026.

TrueCal shows a source label beside each item, such as USDA, label or menu. Its nutrition results remain estimates that can be wrong. Product pages checked October 5, 2026. The guide to calorie trackers with sources looks at what you can inspect beside a food.

Who MyFitnessPal suits

MyFitnessPal may suit someone who wants a very large food database and free food logging, with check marks to identify reviewed entries. Barcode scanning and photo scanning are options for someone willing to pay for Premium.

Someone who wants each number tied to a named source, or mostly records foods without labels, may prefer to look at how other trackers document their food data. The Cronometer source list and MacroFactor common-food database description give places to begin.

Other apps for other needs

Browse other apps people consider instead of MyFitnessPal for a wider list. Read TrueCal side by side with MyFitnessPal if those are the two you are considering.

In TrueCal, you can enter calories manually without paying. Pro covers macros and logging from typed descriptions, voice, photos and barcodes. Pro costs $29.99 a year or $9.99 a month. Eligible customers can try it free for 14 days only in the iPhone and Android apps, where the App Store or Google Play runs the trial. A website subscription charges when you subscribe.

FAQ